Individual bank characteristics that influence bank default empirical evidence from the United States

(1)

Individual Bank Characteristics that influence Bank Default Empirical Evidence from the United States

Genta Bordoniqi 1 August 2014

Abstract

This study investigates the factors that may affect the probability of bank default in the United States. The sample includes banks that are registered with the FDIC in 12/31/1999 and are tracked yearly until 12/31/2013. Predictive variables that are used in a logit estimation procedure are those that incorporate the securitization based intermediation chain role of banks, implied option prices, and individual accounting data. In terms of default, results indicate that there are significant differences among banks of different sizes; the largest of banks do not consistently outperform smaller banks throughout the 14-year tracking period. In addition, evidence suggests of structural alterations in banking before the crises and after the crises. Collateralized mortgage obligations reduced the probability of default during the pre-crises period however, securitized products that that were held until maturity increased the probability of default during the crises. Regulatory-based ratios, such as tier 1 risk-based capital ratio and the total risk based capital ratio are not significant, suggesting the need for regulatory reform. Specifically, Banks that were efficient and at cost of funding advantage contained lower probabilities of default through out the 14-year tracking period.

Universiteit van Amsterdam

Faculty of Economics and Business

Supervisor

(2)

(3)

As more complex financial instruments develop, the role of banks in the financial system is becom-ing increasbecom-ingly intricate. Without proper regulation, such financbecom-ing activities threaten the world economy, as was evident in the recent financial crisis. Banks have a vital role in transmitting funds from savers to borrowers. Therefore, the assessment of a bank’s financial health is a important objective for regulators. Regulators review banks based on on site examinations and off site stress tests. Unfortunately, on site examinations can be quite costly, and the regulator is restricted on the amount of on-site examinations he or she can perform. As a result, they must rely on off-site prediction models. These off-site prediction models consist of statistical models and credit risk models that utilize various bank specific accounting data. By using these statistical models to correctly calculate default probabilities, regulators can distribute their scarce on-site resources to prevent bank defaults that could weaken the stability of the financial system.

Ever since the 1930s, there has been a fair amount of literature on prediction of firm defaults. This naturally progressed into the prediction of bank defaults. The research that has been carried out on bank defaults utilizes bank specific accounting data to explain factors that weaken the competitive position and increase the probability of bank defaults. However, research in bank prediction models has been deficient in regards to the inclusion of the new methods in which banks fund themselves. The banking system has transformed into an ”originate to distribute model”, where banks have the opportunity to rapidly and drastically change their volume of lending - without effecting other components of their asset portfolios.

Understanding how the banking system has evolved from the traditional banking system, to a securities based intermediation chain agent, can reveal the factors that increase or decrease the probability of bank defaults. Whereas literature on bank prediction models has often focused on predictive variables that capture the traditional banking system, the objective of this paper is to construct an accurate bank prediction model utilizing predictive variables that incorporate the new

(6)

domineering, intermediation role of banks.

Before the 1970s, lenders were composed of households and businesses with savings. Borrowers were primarily businesses and households who undertook loans for goods, and home or capital investment. The primary role of the bank was to serve as an intermediary for this transaction (Lut-trell et al , 2012). In the traditional banking system, banks performed three essential activi-ties: maturity transformation, liquidity transformation and credit transformation. Through maturity transformation, banks look to acquire short-term funds and use them to invest in long-term assets. Banks utilize this concept by using short-term deposits, which are considered bank liabilities, to fund loans, which are considered long term liabilities. Naturally, a bank is exposed to interest rate risk, primarily because the maturity of its assets is longer than that of its liabilities. Additionaly, banks face rollover risk and the value of their assets also fluctuates according to the interest rate. Liquidity transformation is similar to maturity transformation and refers to using liquid instruments to purchase less liquid assets. This brings about the fundamental issue - a bank’s assets are not as liquid as it’s liabilities. Furthermore, a majority of banks are required to retain only a fraction of deposits and in a rare occurrence that depositors were to simultaneously withdraw their deposits, the bank would have to resort to fire sales. Consequently, the value of its assets could decrease be-low that of its liabilities. Banks engage in credit transformation by enhancing the credit quality of debt issued. They also use diversification to disperse the risk that is embedded in an individual loan by lending to many different types of borrowers. They earn the difference on the weighted-average rate paid on their liabilities, and the weighted average rate at which they lend. This difference is known as net interest margin.

The modern age of finance has evolved banking from the traditional sense as described above, into a more intricate, wholesale funded, securitization based, intermediation chain. Banks still engage in maturity transformation, liquidity transformation and credit transformation, but through a model known as ”originate to distribute”. In 2007, the assets of securitization pools, composed by firms that issue securities to fund themselves in the United States, were larger than total assets of banks (Adrian and Shin, 2010).

(7)

The manner in which this paper intends to build upon current literature is by using a hazard model to estimate probabilities of default, and incorporating predictive variables that capture the securiti-zation based intermediation chain. Institutions registered with the FDIC in 1999 are tracked yearly until 2013 or until they default to develop a default hazard model. In addition, hazard models are constructed during the financial crisis and pre-crisis to see if structural changes were actually a result of the financial crisis. This study will also include 10,222 FDIC listed institutions that vary in asset size. This will implicitly allow differentiation of institutions into asset classes of less than 100 million, 100 million, up to but not including 1 billion, 1 billion to 10 billion, and greater than 10 billion, to evaluate the performance and vulnerability of different sized institutions. The remainder of this study is arranged in the following manner. Section II provides a thorough de-velopment of firm prediction models and their evaluation in bank default prediction. Section III provides a detailed description of the sample and methodology used. Section IV provides intuition and a description of the predictive variables used, with the specific accounting definitions attached in Appendix I. And finally, Section V is composed of the estimation and analysis of the results obtained. Section VI will provide the conclusion of the study.

(8)

II. Literature Review

The 1930’s was the initial starting point for bankruptcy models utilizing simple ratio analysis. During this time frame, the Bureau of Business Research published a study containing a list of ratios of industrial firms that had failed. This sparked interest on the notion of whether or not failure could be recognized in deteriorating financial performance ratios. It wasn’t until the works of Beaver (1966) and Altman (1968) that the way was paved for the wide range of current research in bankruptcy prediction. Beaver utilized a univariate analysis model, whereas Altman (1968) introduced a multivariate discriminant analysis model. This wealth of literature was followed by models specific to the structure of the firm. Such as in Edminister (1972), whose model was specifically designed for small business failures. Given the importance of banks in the financial system, the issue of bank failure prediction was undertaken in Sinkey (1975) model. Although the use of discriminant analysis was the predominant method up until the 1970’s, it is actually quite restrictive because financial data is assumed to be normally distributed. In addition, it assumes that failed and non-failed banks have an equal variance-covariance matrix, which is not necessarily the case. It was Martin (1977) who made great advancements in these bank prediction models. Martin used a logistic regression and examined all commercial banks that were members of the Federal Reserve System in 1974. The simplicity of the model and superior accuracy in it’s predictions has made it become one of the most widely used models in bank failure prediction.

Models with more predictive covariates do not necessarily contain higher prediction accuracy. Past literature indicates a range of predictors from one to 57 (Bellovary, Giacomino and Akers 2007). Although the logit model has proved its resilience in past literature of bankruptcy prediction mod-els, current studies acknowledge that the logit model is static and does not account for covariates changing with respect to time. Which is an apparent characteristic in factors like macroeconomic data. Shumway (2001) argues that firms change through time and beacuse of the fact that bankrupt-cies are quite rare, researchers must incorporate a larger range of years in order to produce proba-bilities that are not biased or inconsistent. Shumay (2001) proposed that a time hazard model that could be estimated by a multi-period logit model. In his results, this method proved to be superior

(9)

and is the leading model used in survival analysis. To estimate a multi-period logit model, the firm must be tracked each year until an event occurs. In the case of this study, the events are default or the bank is censored. By tracking each firm you are considering the variation over time. Therefore a survival analysis setup produces an unbalanced panel that dramatically increase the sample size. This type of analysis is extended by Glogova et al (2005), who claim that a 2-step logit model is superior because by performing one logit estimation in your initial sample of banks, the researcher assumes that every bank is at risk of defaulting. This may not necessarily be the case in banks that are well-funded. Therefore, by performing a static logit estimation in the first round, you are are able to see the banks that are at risk of defaulting. These banks would then be estimated by a multi-period logit estimation to provide better prediction accuracy. This study had predictive covariates, such as increased lending growth rates and above average deposit rates to capture behavior that would expose the bank to significant risk. The study results in a highly significant covariate, the ratio of profit on ordinary activities to number of employees. This captures management quality and insinuates that when banks are experiencing turbulence, management quality is essential in reducing the probability of default.

Although the models used in past literature vary slightly, the factors that are used to predict bank defaults in literature are mostly based on a CAMEL approach. This approach strives to incorporate different aspects of a banks fragility - such as ratios that relate to capital adequacy, asset quality, management quality, earnings and liquidity. Macroeconomic indicators that control for volatility in the overall economy are included as well. In this study the macoreconomics environment is controlled for by using the S&P 500 VIX. Which is composed of the implied volatilities for a wide range of S&P 500 index options. In his univariate analysis, Beaver (1966) found that net-income to total debt was the ratio that had the highest predictive ability and urged for the incorporation of multiple ratios for further research. Martin (1977) concluded that indicators such as liquid-ity, capital adequacy and earnings where the most prevalent in his study. Consistent with Martin (1977), other studies such as Avery and Hanweck (1984) who performed logit estimations based on the CAMEL approach, also concluded that covariates which were representative of liquidity, capital adequacy and earnings were significant. Barth et al (1985) also concludes, by estimation of

(10)

the static logit model, that the most significant ratios where that of capital adequacy, asset quality and liquidity. These ratios are popular in many studies and include but are not limited to: Net Worth/Total Assets, Interest sensitive Funds/ Total Funds, Net-income/Total Assets and Liquid Assets/Total Assets.

Wheelock and Wilson (2000) estimate a proportional hazards model with 15 predictive covariates. Of which 8 are significant in determining default. These covariates are representative of capital ad-equacy, asset quality, management quality and liquidity. In addition, Arena (2008) study concludes with 8 significant covarites which were representative of capital adequacy, asset quality, manage-ment, earnings and liquidity. Furthermore, Arena (2008) compares the bank failures in East Asia and Latin America, and concludes that even though macroeconomic conditions did have a destabi-lizing effect on the weakest banks, individual bank fundamentals served as the main criterion for increasing default. Kuznetsov (2003) studied the causes of the Russian banking crises in the 1998, and concluded in contrast to Golovan et al(2003), that profitability and liquidity of banks was not related to the probability of failure. Instead, he discovered that firms whos portfolios were exposed to government bonds, had a greater probability in surviving the crises. Although bankruptcy pre-diction models have been used since the 1930s, there has been dramatic structural changes and financial innovation in banking. So much so that the traditional ratios may not properly character-ize the correct degree of risk that a bank has undertaken.

The aforementioned studies were able to achieve sufficient results regarding prediction of defaults, measured by the number of correctly predicited in out sample testing and area under ROC curve, primarily based on financial data unique to the bank and by controlling for the macroeconomic en-vironment. Results of the literature mentioned indicate that distinctive characteristics of the bank that contribute to default can be extracted from reported financial data. However, the landscape of banking has dramatically changed and specific determinates of capital adequacy, asset quality, management quality and liquidity, may not capture the financial health of the bank. According to Adrian and Ashcraft (2012) the traditional forms of financial intermediation, where the bank was exposed to rollover and duration risk, accounted for nearly 100 percent of funding for credit

(11)

intermediation in the 1940s. In recent times this number has now fallen to around 40 percent. The process of securitization has been around for nearly forty years. However, there was a period of rapid securitization before the crisis. In 2005, there was $ 507 billion in mortgage backed securi-tites out of $ 625 billion in sub-prime mortgage loans (Mclean et al, 2010). As a result, the crisis was characterized by dramatic losses on securitized products. A large amount of papers believe that the evolution of securitized products reduced the incentives for banks to properly screen bor-rowers. (Purnanandam 2009) On the other hand, through the process of securitization, banks are able to adjust and partition their risk. According to Juanglu and Pritsker (2008), the process of securitization reduces the risk of insolvency and subsequently increases bank profitability. Further research by Casu et al (2010) suggest that from 2001 to 2007, US bank holding companies that held a greater amount of securitized assets, had composed asset portfolios with lesser credit risk. This study attempts to predict bank defaults by using a hazard model that incorporates a variety of securitization covarities, individual bank accounting information and implied volatilitites of option prices.

(12)

III. Methodology

The survival analysis model used in this study to estimate default probabilities will be a hazard model. In addition, non-parametric estimation such as Nelson-Aalen cumulative hazard estimates, and Kaplan-Meier survival estimates are used for preliminary data analysis. These two forms of non-parametric estimations are used for preliminary data analysis in the Estimation and Analysis section. The Nelson-Aalen cumulative hazard estimates are used to estimate the number of ex-pected bank defaults and give a general idea of the hazard rate shape. In this study, it shows the different hazard rate shapes for different sized institutions. The formula is expressed below, where z is the total amount of entities at risk at tk, xk is the number of defaults at tk

H(t) =

Â

tkt

x_k zk

The Kaplan-Meier survival estimates the survival function for the sample in the study, It measures the length of time banks do not default and used to measure the surivival functions of different sized banks. The formula is expressed below where xkis the number of survivors up to time t and

z_kis the number of defaults.

S(t) =

’

tk<t

xk zk

xk

Default is defined as bankruptcy, acquisition by another institution, or receiving FDIC intervention such as failure assistance transactions, re-openings purchases and assumptions, insured deposit transfers and consignment program institution payoffs from the FDIC. The sample consists of yearly data on commercial and savings institutions, savings banks and savings associations that are registered with the FDIC in 12/31/1999. Theses institutions were tracked until default or until the end of the sample period, 12/31/2013. This considers 10,222 FDIC registered institutions within the 14 year tracking period. During which 4,141 were classified as default, corresponding to the definition of default mentioned above. During the tracking period, each institution has a positive probability of experiencing the event, in this case, default. Occurrence of an event is recorded in intervals with time periods (tk 1,tk]. Periods are indicated by the letter k index, where the kth

(13)

corresponding to each insistitution to assess default probabilities.

A institution can only default once. The occurrence of a default thus becomes inherently condi-tional because a institution can only default in period k if it did not default in period k 1. T is representative of the random variable that designates the specific time period in which default occurs. T is characterized by its conditional probability density function. Where the distribution is dependent upon the probability of default in each time period given that the bank has not defaulted in the previous period. This shape will usually vary and it is dependent on the distribution of risk over time. The prediction period will be one year ahead. hi’s are the central parameters of survival

analysis hazard models, and including a unique set of covariates C = (C1,C2...Cn)for each

insti-tution will individually depict the unique characteristics of each instiinsti-tution in the population. The set of covariates within the hazard definition can be defined as:

h_ik=P[Ti=k|Ti k,C1ik=c1ik,C2ik =c2ik...]

In this study, the proportional hazards model requires a survival analysis data setup. Where every bank in the sample survives through each discrete time period, until the bank experiences the event or is censored by the end of the study. Thus for each commercial bank i, the occurrence of default is recorded through a time variable Zik, keeping in mind that default is a nonrepeatable event. The

proportional hazards model is estimated using a logit model and given the tracking period (1999-2013) the natural logarithm of time to default serves as an appropriate representation in explaining defaults during this tracking period.

Estimation of the hazard model is done through maximum likelihood estimation. Construction of the likelihood function takes into consideration the banks which default in time period k, and is therefore not censored. It also considers banks that do not default during the tracking period and are censored. To construct the likelihood function, both the censored case and the uncensored case must be considered. Based on the derivation of Singer and Willet (1993), for the institution that defaulted in the tracking period, the uncensored observations can be modeled as the probability that they experience the event in time period k, given that they have not experinced it in the previous

(14)

periods k 1.

Pr[Ti=ki] =Pr[Ti=ki|Ti ki]Pr[Ti6= ki 1|Ti ki 1]...

Similarly, the above expression can be reformulated in terms of hazard rates, but for notational matters not including the predictive covarites:

Pr[Ti=ki] = ki 1

’

k=1

hiki(1 hik)

Likewise, considering the situation where default did not occur in the tracking period, the model can be reformulated by mutliplying the conditional probabilities that default did not transpire in the observed time periods. For example periods 1 through ki

Pr[Ti>ki] =Pr[Ti6= ki|Ti ki]Pr[Ti6= ki 1|Ti ki 1]...Pr[Ti6= 2|Ti 2]Pr[Ti6= 1|Ti 1]

Again, if we reformulate the previous equation in terms of hazard rates, excluding for notational purposes, the predictive covararites:

Pr[Ti>ki] = (1 hiki)(1 hiki 1)(1 hiki 2)... = ki

’

k=1

(1 hik)

Given their distinctive covariates, institutions in the sample are assumed to be independent. By im-posing both the uncensored scenario and censored scenario, this produces a log-likelihood function of

n

’

i=1

Pr[Ti=ki]1 piP[Ti>ki]pi

Where the above equation is representative of banks that experience default pi=0(not censored)

and those that did not experience default pi=1(censored). By taking logarithms, the likelihood

function becomes, in terms of hazard rates:

L =

Â

n i=1  (1 pi)log( hiki 1 hiki) + ki

Â

k=1 log(1 hik)

(15)

In addition, we can incorporate the time component through the variable zi j and by further

manip-ulation, the above equation becomes:

L =

Â

n i=1  ki

Â

k=1 zi jlog(_{1 h}hiki iki) +log(1 hik)

By combining like terms and removing the logs we get the likelihood function for the hazard process: L =

’

n i=1 ki

’

k=1 hi jzik(1 hik)1 zik

The hazard model is dependent upon the default sequence and the hazard rate. The hazard model, as proposed by Cox (1972), generalizes that hikcan be reformulated so that the hazard rates have a

logistic dependence on the covariates and time periods because hazard rates are essentially proba-bilities. Therefore, the hazard model becomes:

h_ik= 1

1 + e(Q(t)a1+Cia2)

The hazard model, in relation to the static logit and probit models, has many advantages. Primarily because each entity contains several observations. The survival analysis setup that is required for the hazard model allows you to observe the deteriorating nature of each institution because the data is in institution-year format where each institution is observed until censored or experiences the event. This allows you to incorporate much more data than the static logit and probit models which do not incorporate multiple time periods. In this study fixed effects were not used primarily because of the large number of covariates that attempted to capture heterogeneity between banks.

(16)

IV. Description of Predictive variables

The predictive variables selected for this study are primarily based on previous studies that depict bank vulnerability and literature describing the securitization based intermediation role of banks in the financial system. Accounting definitions of each variable, as defined by the Federal Deposit Insurance Commission, are provided in Appendix I. In this section, predictive variables and their significance to bank default predictability will be described.

Capital adequacy has been a prevalent indicator in numerous studies since Sinkey (1975), Martin (1977), and other recent studies by Derviz and Podiera (2004), and Lanine and Vennet (2005) whom have found relevant predictive variables that are associated with capital adequacy. Capital in banks acts as a safeguard when unexpected losses occur. Therefore, well capitalized banks should have a lower probability of default. In September 1991, the Basel accord defined the amount of core capital that must be held as a percent of risk-weighted assets. Consequently, the amount of risk that a bank is willing to undertake must be funded by the amount of capital as deemed appropriate by the federal regulator. In this study this is represented by the predictive variables rbc1rwa j and rbcrwa j.

Indicators portraying asset quality were deemed as significant in a variety of studies including Gajewki (1988), Martin (1977) and Hayden and Bauer (2004). The asset composition of each of these studies consists of financial instruments that were considered relevant in the specific time period of their studies. This composition made banks vulnerable to characteristics that drove spe-cific crisis in history. For example many banking crises have shown that the composition of an individual bank’s loan portfolio has either allowed them to be impervious to the crisis or increased their probability of default. Riskweighted assets can reflect, the proportion of risky assets that a firm contains, and in this study it is represented by the variable avasset j. In addition, Lis et al. (2000) and Logan (2003) conclude that lending growth of banks can be an indicator for future complications. High lending growth can adversely affect the bank as new borrowers become those that have been declined by other banks Andersen (2008). In this study the magnitude of bad loans

(17)

is measured by the variable pastdueloans. This variable includes loans that were past due by 39 to 89 days and were eventually securitized and sold. Variables such as nclnlsr and ntlncsr measure the magnitude of writeoffs.

Asset securitization is the process of pooling together numerous debt obligations composed of mortgages, credit card receivables and auto loans into one security. Furthermore, these security can be fragmented into different pieces and sold as bonds or other securities. Each fragment then has the risk profile of the entire group, not just one obligation. Risk profiles of each fragment differ based on credit quality of the securities pulled together. The reasons why banks undergo securiti-zation activities are highly correlated and difficult to isolate. Acharya, Schnable and Suarez (2011) conclude that banks undergo securitization for regulatory arbitrage purposes. Regulation was put in place, which stated that assets merged onto balance sheets from asset-backed commercial paper conduits would not need to be included in risk based capital ratios, but would have to be used as a ten percent credit conversion factor. This incentivized banks to structure liquidity enhancing guar-antees to reduce their regulatory capital (Adrian and Ashcraft 2012). Literature on the effect of securitization is mixed. Pennachi, (1988), (Gorton and Soules, 2006) conclude that securitization allows banks to reduce risk because it inadvertently allows banks to reduce regulatory capital costs and bankruptcy costs. Others indicate that banks originated loans for the sole purpose of collecting fees when selling the loans (Rosen, 2010). Alternative points of view suggest that securitization allowed banks to reduce risk, predominately in home mortgages, because they are able to lend to riskier borrowers and sell those risky loans. This enables them to hold less risky loans (Cebenoyan and Strahan, 2004). It is important to note the difference between banks that sell loans and banks that securitize assets. A multitude of banks sell loans, but only a small fraction of banks securitize assets. By securitizing assets the bank is creating the pools of assets that are sold. This activity is usually only done by large bank holding companies. Whereas many banks of all sizes primarily sell loans. In my sample, the percentage of banks that conduct securitization activities is 2.1 %, whereas almost all sell loans. In this study, credit enhancements and securitization activities are captured by enhacsec, subsec, retsec and totalsec.

(18)

Mortgages used as collateral in securitization, known as MBS, are originated by either government sponsored enterprises (GSE) or private firms such as commercial banks. Many of the mortgages that are sold, end up with GSEs and can be sold many times before they end up in the asset pool. As mentioned before, a small percentage of banks, mostly large holding companies actually securitize these mortgage pools, but almost every bank sells mortgages that ultimately end up in MBS pools. Although there are many ways in which a mortgage ends up in these pools, the most direct method is when the bank themselves pools the mortgages and originates the MBS using the loans as collateral. Part of the sale process is holding the loan until it sells because it is not plausible to sell the loan the same day you originate it. Therefore, there is inherent risk in the originate to distribute model because a change in mortgage rates while the loan is being held can create either a gain or loss to the bank. Moreover, securitization markets are inherently risky because they can shut down. As we saw in the recent financial crisis, banks were ultimately left holding the loans that they initially intended to sell. In this study, collateral debt obligations and mortgage backed securities are represented by variables such as scmtgbk, idsccmo and idsccmt.

Because a banks liabilities are composed of short-term deposits, and their assets are composed of loans, banks are naturally at risk of contagion. Contagion is often defined as systematic risk, where complications that result in one bank spillover to other banks. Even if the banks did not have any characteristics in common. If a bank is experiencing financial difficulties it can effect depositors perceptions, of other banks and influence others to withdraw their deposits. Early works such as Diamond and Dybvig (1983) develop the concept of multiple equlibria and conclude that bank runs occur because depositors shift their expectations based on the information that they observe. Bank runs have been modeled thoroughly in literature and they form the fundamental notion that depositors will try and avoid liquidation by withdrawing their deposits early - solely because they believe that others will do the same. Ideally, to empirically assess systematic risk you would need interbank exposures, but that is difficult to obtain. Therefore, I will assess systematic risk with variable idtrcomb which is representative of the amount of deposits of other commercial banks and other depository institutions that a bank posses. Keeping the limitations in mind, this variable does not explicitly show which bank an institution is mostly exposed to, but it does show the extent

(19)

to which an institution is exposed to the banking sector. Therefore, if an asymmetric shock had adverse effects on the banking sector then banks holding a large portion of deposits from other banks will be even more likely to experience defaults. Furthermore, predictive variables such as lndepac and lndepcb represent loans to depository institutions, and acceptances and loans to commercial banks in the U.S., respectively. These variables disclose to what extent each bank is exposed to the banking sector in the United States. Insinuating that the level of riskiness of the institutions would be dependent upon the health of the banking sector.

Concentrating loan portfolios can leave banks vunerable to asymmetric shocks. A bank extends loans to many types of individuals, however if it does so to affiliated companies, dominant share-holders, executive officers, subsidiaries, and members of the board of directors, it exposes the bank to insider loans were the opportunity for abuse arises. Questionable lending practices have been a characteristic of many banking crises, including the recent financial crisis and therefore it is imperative to assess the implications that insider loans give raise too. According to Kummer and Arshadi (1989) managers that have a limited liability to the bank have an incentive to come together with insider borrowers to circumvent the shareholders and regulators. It is possible that a bank grants advantageous contract terms to insider loans as compared to the loan it would grant to a non-insider borrower. If this type of lending is practiced at considerable levels, it could increase the likelihood of default. As insiders whom have a limited liability to the bank, have incentives to engage in behavior of selfinterest, rather than for the benefit of the institution. Since this poses a threat, predictive variable lnexamt capture a banks extension of credit to all of its executive officers, directores, principal shareholders and their related interests.

(20)

V. Estimation and Analysis Graph 1:

Non-parametric estimation, such as the Kaplan-Meier, reveals differences in survival estimates solely based on the size of the institution. Less100mil stands for institutions that have adjusted-average assets of less than 100 million, greater100millto1bill stands for institutions that have adjusted-average assets of 100 million up to but not including 1 billion, onebilltenbill stands for institutions that have an adjusted average assets of 1 billion and up to 10 billion, and large stands for institutions in which adjusted-average assets are greater than 10 billion. It is not surprising to see banks of different size have varying survival rates. As smaller banks in the United States have often quite different business models then that of large bank holding companies. Large bank holding companies include derivative market making, wholesale banking and capital markets – quite a diverse assortment of services in comparison to smaller banks that primarily focus on maturity transformation. Large fixed costs impede smaller banks from entering in to these diverse financial specialties.

There are pros and cons associated with large holding companies; medium sized financial institutions and small local banks. As such, it is difficult to determine, in terms of default, the implications that the size of the financial institution creates. Needless to say that since the 1980’s, there has been an enormous amount of consolidation of small banks to form large

(21)

holding companies - especially since the late 90’s and during the recent financial crisis.1 We tend to think of larger financial institutions as being better in the sense that they operate in more sophisticated markets and have access to greater capital. However, the preceding graph shows the survival estimates of different sized banks through a fourteen-year tracking period. The largest financial intuitions that have assets greater than 10 billion do not consistently outperform institutions that have less than 100 million. In fact after 2009 (after period 10 in both graph 1 and 2), in the midst of the financial crisis, institutions that had less than 100 million in adjusted average assets had a higher probability of surviving.

Graph 2:

This suggests that even large holding companies are vulnerable to financial shocks. Nonetheless, the recent financial crisis involved sophisticated financial instruments that made large holding companies more vulnerable than institutions containing adjusted average assets of less than 100 million. Primarily because those institutions did not hold such toxic assets. However, maybe it is not fair to compare the superiority of bank sizes in terms of a tail event. If we observe graph 1 and 2 before the financial crisis we see that smaller institutions had a lower survival probability then that of large holding companies. This implies a common notion among individuals, but why do we necessarily believe that bigger is better? When measuring the effectiveness of any !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

1_{FDIC Community Banking Study, December 2012} 2_{FDIC Community Banking Study, December 2012}

(22)

institution in terms of size, we tend to think of the concept of economies of scale. If a firm increases its output and the average cost of producing the additional unit of output decreases, then the firm is experiencing increasing economies of scale.

Banking is predominantly a service-orientated industry. So conceptually, it is difficult to define and measure a unit of output. Jacewitz and Kupiec (2012) measure output as the sum of interest expenses, provisions for loan and lease losses, and non-interest expenses, divided by balance sheet assets. Their study concludes that economies of scale provide no significant benefits after 500 million in asset size. This result coincides with graphs 1 and 2, as institutions with medium average assets have a higher survival rate through out the 14-year tracking period.

Another common performance measure of banks in terms of size is the concept of efficiency. It is the measure of non-interest expenses divided by the sum of net interest income and noninterest income. Therefore banks that have higher efficiency ratios are less efficient. Graph 3 depicts the diversion of efficiency ratios of institutions of differing asset size.

Graph 3: Quarterly Efficiency ratios 0.0%! 10.0%! 20.0%! 30.0%! 40.0%! 50.0%! 60.0%! 70.0%! 80.0%! 90.0%! 100.0%! 84:1! 85:4! 87:3! 89:2! 91:1! 92:4! 94:3! 96:2! 98:1! 99:4! 01:3! 03:2! 05:1! 06:4! 08:3! 10:2! 12:1! 13:4! Assets!>!$10!Billion! Assets!$1!Billion!:!$10! Billion! Assets!$100!Million!:!$1! Billion! Assets!<!$100!Million!

(23)

Smaller institutions are less efficient than larger institutions. The greatest non-interest expense in institutions of all asset groups is employee-associated costs.2 This is an essential component in explaining the efficiency gap seen between larger and smaller institutions because as Graph 4 shows, larger institutions have experienced higher productivity gains. In addition, Graph 5 shows that overtime, non-interest expenses have increased for all different asset size institutions. However, at the same time, larger institutions experienced increases in productivity gains, which have offset the increase in non-interest expenses. Smaller institutions did not benefit from productivity gains and therefore have had higher efficiency ratios.

Graph 4: Average Assets Per Employee ($ Millions)

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

2_{FDIC Community Banking Study, December 2012}

$0.000 $1.000 $2.000 $3.000 $4.000 $5.000 $6.000 $7.000 $8.000 $9.000 84: 1 85: 3 87: 1 88: 3 90: 1 91: 3 93: 1 94: 3 96: 1 97: 3 99: 1 00: 3 02: 1 03: 3 05: 1 06: 3 08: 1 09: 3 11: 1 12: 3 14: 1 Assets > $10 Billion Assets $1 Billion - $10 Billion Assets $100 Million - $1 Billion Assets < $100 Million

(24)

Graph 5: Quarterly Non-Interest Expense as a % of Average Assets

In addition, if we observe the quarterly cost of funding earning assets (Graph 6), which is the total interest expense paid as a percentage of assets, we see that before the financial crisis, larger institutions were at a slight cost disadvantage. After the financial crisis smaller banks no longer enjoyed the cost of funding advantage and this contributes to the difference in net interest margins between smaller banks and larger banks (Graph 8).

Graph 6: Quarterly Cost of Funding Earning Assets

0.00% 0.50% 1.00% 1.50% 2.00% 2.50% 3.00% 3.50% 4.00% 4.50% 84: 1 85: 3 87: 1 88: 3 90: 1 91: 3 93: 1 94: 3 96: 1 97: 3 99: 1 00: 3 02: 1 03: 3 05: 1 06: 3 08: 1 09: 3 11: 1 12: 3 14: 1 Assets > $10 Billion Assets $1 Billion - $10 Billion Assets $100 Million - $1 Billion Assets < $100 Million 0.0% 20.0% 40.0% 60.0% 80.0% 100.0% 120.0% 140.0% 84: 1 85: 4 87: 3 89: 2 91: 1 92: 4 94: 3 96: 2 98: 1 99: 4 01: 3 03: 2 05: 1 06: 4 08: 3 10: 2 12: 1 13: 4 Assets > $10 Billion Assets $1 Billion - $10 Billion Assets $100 Million - $1 Billion Assets < $100 Million

(25)

Graph 7: Quarterly Loss Provision as % of Net Operating Revenue

In conclusion, non-parametric estimates of survival and hazard rates reveals different risk estimates depending on asset size of the institution. Further investigation reveals that larger institutions have benefitted from productivity gains and higher levels of efficiency that allowed them to obtain higher survival rate estimates before the financial crisis. During the financial crisis, larger institutions were predominately exposed to toxic assets and therefore suffered unprecedented losses (as can be seen Graph 7). Whereas smaller institutions that were not exposed to such assets, because of the high fixed costs that the securities and derivative market making markets require, contained higher survival estimates. Furthermore, before the financial crisis smaller institutions had a minor cost advantage that they lost through the financial crisis. Since net interest income is composed of interest expense, the cost of funding advantage allowed smaller institution to surpass larger institution in terms of net interest margins. However, smaller institutions lost the cost of funding advantage through the financial crisis and are now operating at moderate cost disadvantage. Through this preliminary analysis, institutions that had average assets greater than 100 million, up to but less than 1 billion were more resilient before the crisis and during the crisis.

0.00% 2.00% 4.00% 6.00% 8.00% 10.00% 12.00% 84: 1 85: 3 87: 1 88: 3 90: 1 91: 3 93: 1 94: 3 96: 1 97: 3 99: 1 00: 3 02: 1 03: 3 05: 1 06: 3 08: 1 09: 3 11: 1 12: 3 14: 1 Assets > $10 Billion Assets $1 Billion - $10 Billion Assets $100 Million - $1 Billion Assets < $100 Million

(26)

Graph 8: Quarterly Net Interest Margins

This section examines the logit estimations within the entire tracking period, pre-crisis and during the crisis. In each case, the prediction length is one year. Table 1, displays the initial logit estimation of the entire tracking period (1999-2013), including only predictive variables significant at the 10% level. Of the 33 predictive variables included in the initial estimation, 12 were deemed significant at the 10% significance level. Table 2 displays the logit estimation for the pre-crisis period (1999-2006). Of the initial 33 predictive variables, 13 were significant at the 10% significance level. Table 3 displays the logit estimations for the crisis period (2007-2013). Of the initial 33 predictive variables, 9 were significant at the 1% significance level. It is important to note that all of the predictive variables are taken as a percentage of adjusted average assets. Tables 1, 2 and 3 contain logit estimations with significant predictive variables that are classified under the CAMEL approach and ratios that demonstrate the core activities of a bank. All of the models contain a S&P 500 VIX control variable to control for the state of the economy and a S&P 500 VIX variable that interacts with banks that have average adjusted yearly assets of less than 100 million.

0.00% 0.50% 1.00% 1.50% 2.00% 2.50% 3.00% 3.50% 4.00% 4.50% 5.00% 84: 1 85: 3 87: 1 88: 3 90: 1 91: 3 93: 1 94: 3 96: 1 97: 3 99: 1 00: 3 02: 1 03: 3 05: 1 06: 3 08: 1 09: 3 11: 1 12: 3 14: 1 Assets > $10 Billion Assets $1 Billion - $10 Billion Assets $100 Million - $1 Billion Assets < $100 Million

(27)

Table 1: Logit Estimation of 1999-2013

Default Coef. Std. Err. z P>z [95% Conf. Interval]

nclnlsr 0.099 0.005 19.050 0.000 0.089 0.110 idtrcomb 0.606 0.229 2.650 0.008 0.158 1.054 roa -44.390 5.321 -8.340 0.000 -54.819 -33.961 sdep 0.545 0.320 1.710 0.088 -0.081 1.172 idsccmo 0.640 0.304 2.100 0.035 0.043 1.236 voliab 0.988 0.153 6.460 0.000 0.688 1.288 lntime -2.873 0.036 -80.210 0.000 -2.944 -2.803 less100mill#c.vixclose 0.188 0.006 29.790 0.000 0.175 0.200 greater100mil~l 4.914 0.150 32.810 0.000 4.621 5.208 onebilltenbill 5.125 0.162 31.670 0.000 4.808 5.442 large 5.138 0.195 26.290 0.000 4.755 5.521 vixclose -0.046 0.005 -9.890 0.000 -0.055 -0.037

It is interesting to note that predictive variables that assessed change specifically in loans (spastdueloans) to alert regulators of credit risk were not significant. A possible explanation that coincides with Anderson (2008) could be scenarios of adverse selections. Adverse selection occurs when loan growth is exceedingly high. High loan growth can indicate that borrowers with poor credit quality were taken into consideration in the selection process, especially if the growth is significantly higher than the mean of the overall industry. This could suggest that a non-linear relationship is more appropriate.

Return on assets (ROA) is significant and is representative of the earnings factor in the CAMEL approach. Having a strong earnings base is essential in banking but also quite difficult because earnings are sensitive to volatile market interest rates. In addition, to retain viable returns, banks have to refrain from nonrecurring events and extraordinary events that can reduce the quality of their earnings. Table 1 suggests that a higher return on assets reduces the probability of defaults - higher return on assets signifies higher efficiency. This outcome was at the expense of return on equity because both variables are highly correlated and could not be included in the regression at the same time. By omitting return on assets and including return on equity in the regression, return on equity is not significant. This coincides with several studies such as Anderson (2008), Arena (2008) etc.

(28)

Predictive variables such as tier 1 risk-based capital ratio and total risk based capital ratio are not significant in any of the models. A reason for this could be the regulatory arbitrage argument. As previously stated, only a small fraction of banks actually securitize products, in this sample 2.1%. But a majority, as in this sample 80% of banks, sells loans. According to Elul 2005, banks are able to reduce their capital requirements by selling loans. Therefore implying that such measures that assesses capital as a percent of risk weighted assets are not appropriate in identifying the risk that an institution is undertaking - given that the risk–weighted asset measurement can be manipulated. Calomiris and Mason (2004) In addition, variables that measured credit enhancements were not found significant as only a small percentage of banks actually engage in such activities.

lnexamt -3.041 1.529 -1.990 0.047 -6.037 -0.044 idsccmo -1.876 0.535 -3.510 0.000 -2.924 -0.827 eeffr 0.193 0.033 5.870 0.000 0.129 0.258 roa -8.133 4.564 -1.780 0.075 -17.078 0.811 snim 15.699 3.288 4.780 0.000 9.256 22.143 lntime -3.337 0.048 -69.740 0.000 -3.431 -3.243 less100mill#c.vixclose 0.234 0.010 23.180 0.000 0.215 0.254 greater100millto1bill 5.572 0.220 25.290 0.000 5.140 6.004 onebilltenbill 5.945 0.228 26.120 0.000 5.499 6.391 large 6.210 0.260 23.890 0.000 5.700 6.719 vixclose -0.194 0.007 -27.700 0.000 -0.208 -0.181

At first glance, the positive sign on the variable sdep might seem counterintuitive, however sdep measures the standard deviation in the sum of all deposits. Banks that are facing liquidity issues are inclined to raise their deposit rates above their competitors to attract funds. Therefore, significant increases in the amount of deposits over a year can signal liquidity issues and a decrease in deposits undercuts the banks major source of funding. Derviz and Podiera (2004) Volatile liabilities in a consolidated basis are represented by the predictive variable voliab. In Table 1, which considers the entire tracking period, an increase in the level of volatile liabilities, such as federal fund purchases and securities sold under agreements to repurchase, increases the

(29)

probability of default. When we initially consider the composition of volatile liabilities (Appendix I), they may seem as an important source of funding. However, these items are highly contingent on factors that the bank does not control, and are not a stable source of funding.

Higher efficiency ratios as previously mentioned means that banks are less efficient. Pre-crisis banks with higher efficiency ratios had a higher probability of defaulting (Table 4). During 1999 to 2006, smaller banks experienced higher efficiency ratios predominately due to a lack of productivity gains. This coincides with the findings of Jacewitz and Kupiec (2012). Table 3 shows that during the crisis period, higher intexpy increases the probability of default during the crisis (Table 3). Higher intexpy leads to a higher cost of funding earning assets meaning that the interest that they pay on deposits and on other borrowed money is high therefore reducing net interest margins. Furthermore, higher net income reduced the probability of default during the crisis period. This is expected as higher income creates a cushion in which shocks can be absorbed. Higher standard deviation in net interest income increases default. Banking is an industry in which stability is rewarded, and interest income volatility suggests that the bank is relying on nonrecurring events or extraordinary gains for self-funding. Which is not sustainable in the long run. Further, in all models, the longer the survival time of a bank, the lower the probability of default. As anticipated, experience within an industry increases viability.

Collateralized mortgage obligations that were held at maturity that were issued or guaranteed through FNMA or FHMLC increased the probability of default during the entire tracking period and the crisis. However when investigating the logit estimation for the pre-crisis period individually, collateralized mortgage obligations reduced the probability of default. Before the financial crisis, collateral mortgage obligations were profitable. Especially those guaranteed by FNMA, FHLMC or GNMA because the economy was going through a refinancing boom and selling these obligations was quite easy up until the crisis when the securities market shut down. Rosen (2010) suggests that banks increase their volume of lending during refinancing booms and therefore lend to riskier borrowers. This result coincides with that of Cebenoyan and Strachan (2004) who conclude that commercial loan sales do allow banks to make riskier loans, but ultimately hold less risk than banks that hold such securities because they are able to sell the riskier loans. However, during the crisis period the securities market was essentially shut down,

(30)

leaving banks to hold those risky securitized loans. Thus banks that held their mortgage obligations increased their probability of default whereas periods in which banks were able to sell their securitized mortgages, reduced their risk.

nclnlsr 0.074 0.006 11.570 0.000 0.061 0.087 idtrcomb 0.875 0.323 2.700 0.007 0.241 1.509 lndepcb 8.529 2.819 3.030 0.002 3.003 14.055 snim 25.878 4.399 5.880 0.000 17.256 34.501 idsccmo 3.263 0.410 7.950 0.000 2.458 4.067 lnexamt 7.389 2.017 3.660 0.000 3.436 11.342 netinc -30.271 1.965 -15.410 0.000 -34.121 -26.420 intexpy 16.463 3.378 4.870 0.000 9.843 23.083 lntime -5.479 0.216 -25.370 0.000 -5.902 -5.055 less100mill#c.vixclose 0.331 0.019 17.000 0.000 0.293 0.370 middlesize 9.698 0.595 16.300 0.000 8.532 10.865 bigsize 9.898 0.596 16.600 0.000 8.729 11.066 vixclose 0.012 0.007 1.850 0.065 -0.001 0.025

Loans and leases that were 90 days past due are statistically significant at the 1% level during the entire tracking period – but more specifically, during the financial crisis. Loans are the essential components of the fundamental business model of all banks. Therefore, it is no surprise that high loan write-off increases the probability of default. This result coincides with that of Gloggova and Halling et al (2005). In addition, predictive variables such as lndepcb show the total loans that a bank makes to commercial banks in the United States. As Table 3 shows, a higher concentration of loans to other commercial banks as measured by lndpcb increases the probability of default. This variable was only significant at the 1% significance level in the model that included the crisis period. Having a concentrated lending portfolio in any sector is always detrimental, but banks that lend to other commercial banks are more vulnerable to counterparty risk. This was more apparent during the financial crisis then in the pre-crisis period. It is interesting to note that insider loans represented by lnexamt increased the probability of default during the crisis and decreased the probability of default before the crisis. This

(31)

corresponds with the argument that loan concentration as lnexamt represents loans to shareholders and their related interests. Which were more vulnerable during the financial crisis.

The dummy variables that designate the different asset classes are significant at the 1% significance level in all the models. The interaction between banks of less than 100 million and the VIX is positive indicating that keeping other factors equal, small banks are more vulnerable to default than larger banks in bad states of the economy. Such interactions are not significant with other differing asset amount dummies. Other asset amount dummies are significant at the 1% significance level and positive, but as the at-means regressions in the logit estimations show in Appendix III, keeping other factors constant, larger institutions had a lower probability of default than institutions with adjusted average yearly assets of less than 100 million in all models, pre-crisis, during the crisis and through the entire tracking period. However, based on the logit estimations there is no clear distinction between institutions that have an adjusted average amount greater than 100 million.

The out of sample testing consisted of a 20% random sample where the predictive performance of the models is based on the Receiver Operating Characteristic (ROC) curve. The area under the ROC curve represents the probability that the specific model assigns a higher probability of default to institutions that have defaulted than to institutions that have not defaulted. Therefore when the area underneath the curve is equal to 0.50 the model has no predictive performance. When the area underneath the curve equals 1, the model can perfectly distinguish between defaulted banks and non-default banks. Figure 1 displays the respective ROC curve for each model. The respective areas under the curve are 0.82, 0.83 and 0.93, where according to Tape (2006) the values of the area under the curve from 0.50 to 0.60 are categorized as a “fail”, 0.60 to 0.70 are classified as “poor”, 0.70 to 0.80 are classified as “fair”, 0.80 to 0.90 is classified as “good” and 0.90 to 1 is classified as “excellent”.

(32)

Crisis 2007-2013 Pre-Crisis 1999-2006

(33)

VI. Summary and Conclusion

Many papers have assessed bank vulnerability and have attempted to capture the strength of the institution through financial ratios. Papers such as Anderson (2008), Glogova and Halling et al (2005), Wheelock, D.C, and Wilson (2000), Estrella and Park et al (2000) etc. Although there were varying samples and results, these papers concur that defaults are predictable by

incorporating individual institution characteristics. I extend these papers in several ways. First, my sample consists of 10,222 commercial banks, savings associations and thrift institutions in the United States, with a tracking period of 1999 to 2013. The large sample size allowed for a more in-depth investigation of the structural differences of different asset sized banks. In

addition, the tracking period is composed of pre-crisis and financial crisis years, this allowed for the opportunity to observe the performance of banks under unique circumstances. Finally, I include predictive variables that incorporate the securities-based intermediation chain role of banks.

Results indicate that, in terms of vulnerability to default, there are significant differences of banks of different size. The largest banks, those that have adjusted average assets greater then 10 billion, did not consistently outperform in terms of survival estimates during the financial crisis. Unlike the larger banks, smaller institutions were not significantly exposed to securitization activities as these operations have immense fixed costs. In pre-crisis years, where financial distress was not present, institutions that had an adjusted average asset of less than 100 million faced a lower probability of surviving. Large banks operate in many sophisticated markets that smaller banks do not have access to. However if we consider the fundamental banking operations between the two, we notice sizable differences that leave smaller institutions at a disadvantage. Smaller institutions are less efficient then larger institutions and larger institutions also have larger net interest margins. Less efficient banks had a higher probability of default during the pre-crisis period. In addition, banks that were at operating at a cost disadvantage operated with a higher interest expense and this increased their probability of default.

Securitized products such as collateralized mortgage obligations reduced the probability of default during the pre-crisis period. On the other hand, securitized products that were held at

(34)

maturity increased the probability of default during the crisis and in the model that encompassed the total tracking period. This suggests that securitization allows banks to dramatically change their volume of lending. However, in periods where the securitization markets are closed, banks have to hold loans they were initially thought they were going to sell. These types of loans are composed of riskier borrowers. Rosen (2010)

The pre-crisis model and during crisis model differ in the significance of ratios indicating the structural alterations between both periods. Composition of the regulatory framework, economic environment, and banking framework will change eventually making specific predictive variables outdated as new financial shocks develop. However, ratios that focus on banks’ core business operations, such as efficiency, profitability and cost of funding, are to be significant in all the models. A limitation to this empirical analysis, given the vast amount of banks in the sample, is that it is difficult to distinguish which banks defaulted due to poor financial health or were bought out for consolidation purposes - not necessarily because of eroding financial health. The prediction accuracy for the out of sample estimation is high and therefore the altered definition is disregarded but still noted.

(35)

Appendix 1

!

Adjusted Average Assets for Leverage Capital

Purposes

Avassetj

Average assets-adjusted is based on the risk-based capital definitions for prompt corrective action (PCA) and includes:

Average assets (from schedule RC-K) - disallowed intangible assets - unrealized loss on equity securities - unrealized holding gains or losses on available-for-sale securities according to FASB 115 - adjustments for financial subsidiaries (Beginning in March 2001)

Average Assets Assetfive

Year-to-date average of the total assets represented on the balance sheet. Used as the denominator for year-to-date income as a percent of average assets. The number of quarterly values used in the calculation depends on the date of the data. December reporting period - (Previous December assets + March assets + June assets + September assets + December assets) / 5

Bank Assets Sold and

Securitized totalsec

1-4 Family Residential Loans Bank Securitization Activities Outstanding Principal Balance Of Assets Sold And Securitized With Servicing Retained Or With Recourse Or Other Seller-Provided Credit Enhancements - 1-4 Family Residential Loans + Home Equity Lines Bank Securitization Activities Outstanding Principal Balance Of Assets Sold And Securitized With Servicing Retained Or With Recourse Or Other Seller-Provided Credit Enhancements -Home Equity Lines + Credit Cards Receivables Bank Securitization Activities Outstanding Principal Balance Of Assets Sold And Securitized With Servicing Retained Or With Recourse Or Other Seller-Provided Credit Enhancements -Credit Card Receivables + Auto Loans Bank Securitization Activities Outstanding Principal Balance Of Assets Sold And Securitized With Servicing Retained Or With Recourse Or Other Seller-Provided Credit Enhancements -Auto Loans + Other Consumer Loans Bank Securitization Activities Outstanding Principal Balance Of Assets Sold And Securitized With Servicing Retained Or With Recourse Or Other Seller-Provided Credit Enhancements - Other Consumer Loans + Commercial & Industrial Loans Bank Securitization Activities Outstanding Principal Balance Of Assets Sold And Securitized With Servicing Retained Or With Recourse Or Other Seller-Provided Credit Enhancements -Commercial And Industrial Loans + All Other Loans and All Leases Bank Securitization Activities Outstanding Principal Balance Of Assets Sold And Securitized With Servicing Retained Or With Recourse Or Other Seller-Provided Credit Enhancements -All Other Loans And All Leases.

Collateralized Mortgage

Obligations idsccmo

Mortgage-backed securities (CMOs and REMICS) held-to-maturity at amortized cost and available-for-sale at fair value which are either issued or guaranteed through FNMA or FHLMC, or privately issued and collateralized by mortgage-backed securities issued or guaranteed by FNMA, FHLMC, or GNMA and all other privately-issued. This includes securities held in trading accounts and privately issued collateralized mortgage obligations (REMICS)

Commercial

Mortgage-Backed Securities idsccmt

Commercial mortgage pass-through securities and other commercial mortgage-backed securities (issued by U.S. government-sponsored agencies or by others) on a consolidated basis.

Cost of Funding Earning

Assets Intexpy

Annualized total interest expense on deposits and other borrowed money as a percent of average earning assets on a consolidated basis.

Efficiency Ratio Eeffr Noninterest expense, less the amortization expense of intangible assets, as a percent _{of the sum of net interest income and noninterest income.}

Held to Maturity Securities

(book value) scha

Total securities designated to be held to maturity, reported at amortized cost (book value).

(36)

Insider Loans Lnexamt The aggregate extension of credit at the reporting entity(ies) to all the executive _{officers, directors, principal shareholders and their related interests.}

Loans to Commercial

Banks in U.S. Lndepcb

Total loans to commercial banks located in the U.S. and acceptances of such banks. This item is not reported by institutions with less than $300 million in total assets.

Loans to Depository Institutions and

Acceptances

Lndepac

All loans (other than those secured by real estate), including overdrafts, to banks, other depository institutions, and other associations, companies, and financial intermediaries whose primary business is to accept deposits and to extend credit for business or for personal expenditure purposes. Also the bank’s holdings of all bankers acceptances accepted by other banks that are not held for trading. Acceptances accepted by other banks may be purchased in the open market or discounted by the reporting bank.

Maximum Amount of

Credit Exposure Retained Retsec

1-4 Family Residential Loans Maximum Amount Of Credit Exposure Arising From Recourse Or Other Seller-Provided Credit Enhancements Provided To Structures Reported In RC-S, Item 1. In The Form Of: Retained Interest-Only Strips (Included In The Securities Schedule And/Or Other Assets)- 1-4 Family Residential Loans + Home Equity Lines Maximum Amount Of Credit Exposure Arising From Recourse Or Other Seller-Provided Credit Enhancements Provided To Structures Reported In RC-S, Item 1. In The Form Of: Retained Interest-Only Strips (Included In The Securities Schedule And/Or Other Assets)- Home Equity Lines + Credit Cards Receivables Maximum Amount Of Credit Exposure Arising From Recourse Or Other Seller-Provided Credit Enhancements Provided To Structures Reported In RC-S, Item 1. In The Form Of: Retained Interest-Only Strips (Included In The Securities Schedule And/Or Other Assets)- Credit Card Receivables + Auto Loans Maximum Amount Of Credit Exposure Arising From Recourse Or Other Seller-Provided Credit Enhancements Seller-Provided To Structures Reported In RC-S, Item 1. In The Form Of: Retained Interest-Only Strips (Included In The Securities Schedule And/Or Other Assets)- Auto Loans + Other Consumer Loans Maximum Amount Of Credit Exposure Arising From Recourse Or Other Seller-Provided Credit Enhancements Provided To Structures Reported In RC-S, Item 1. In The Form Of: Retained Interest-Only Strips (Included In The Securities Schedule And/Or Other Assets)-Other Consumer Loans + Commercial & Industrial Loans Maximum Amount Of Credit Exposure Arising From Recourse Or Other Seller-Provided Credit Enhancements Provided To Structures Reported In RC-S, Item 1. In The Form Of: Retained Interest-Only Strips (Included In The Securities Schedule And/Or Other Assets)- Commercial And Industrial Loans + All Other Loans and All Leases Maximum Amount Of Credit Exposure Arising From Recourse Or Other Seller-Provided Credit Enhancements Provided To Structures Reported In RC-S, Item 1. In The Form Of: Retained Interest-Only Strips (Included In The Securities Schedule And/Or Other Assets)-All Other Loans, Leases and Other Assets.

Individual bank characteristics that influence bank default empirical evidence from the United States

Contents

Â

’

’

’

’

Â

Â

Â

Â

’

’