Credit Risk Transfer Instruments in U.S. Commercial Banks

(1)

Commercial Banks

---- Before Crisis, During Crisis and Recession

Name: X.CHEN

Student Number: S2234122

Study: Master of Finance

Supervisor: Prof. K.F. Roszbach Faculty: Economics and Business University: University of Groningen

Date: Aug. 10, 2013!

Abstract

Many banks experienced peaks and troughs in the first decade of the 21st century, covering the Subprime Crisis erupted in August 2007. Credit risk transfer (CRT) instruments, including loan sales, securitizations, and credit derivatives, are debatable on the role played in this crisis. This thesis investigates incentives of CRT instruments and impact of CRT activities on risks of U.S. commercial banks before-crisis, during crisis and recession. Banks are exposed to higher risks and more frequently to default when they are more active in selling and securitizing loans and trading credit derivatives, since CRT instruments played the role of funding tool over risk transferor in this empirical study. Moreover, the study suggests a more influential role played by loan sales and securitizations than credit derivatives.

Jel code: E44, G01, G21

(2)

!

1. Introduction

In the first decade of the 21st century, banks used credit risk transfer (CRT) tools to control credit risks level. Loan sales, loan securitizations, and credit derivatives are common CRT instruments. Loan sales and loan securitizations are similar in selling the future payment streams generated from underlying loans to other institutional or private investors. But loan securitizations are achieved through a special purpose vehicle and new securities issuance, rather than direct loan sales. Unlike loan sales and loan securitizations, credit derivatives are contracts between banks and other investors. These investors trade derivatives with banks to insure the bank a payment in case of default. Hence, these CRT activities enable capital relief of banks, and sometimes create cash inflows by loan sales / securitizations.

Traditional loan programs of banks are held on the balance sheet until maturity or default, and the bank’s risk are mainly managed by constructing well-diversified portfolios (Chiesa, 2008). In order to raise more funds, banks search for fund resources besides the traditional ones. By trading loans and credit derivatives, banks are now engaging extensively in loan sales and more generally in CRT activities (Minton et al., 2005). However, the engagement in CRT activities is debatable. On one hand, Buffett (2002) believes that CRT activities may harm the stability of the financial sector. While on the other hand, Greenspan (2005) pointed out the evidence that in the early years of 2000s, the U.S. financial sector as a whole was not harmed by recession, because of the extensive use of CRT instruments. He argues that loan securitization and credit derivatives were particularly important. But as the use of the CRT instruments kept increasing in the first decade of the 21st century, was the financial systems really unharmed?

(3)

financial crisis ever since the Great Depression in 1930s by many economists. Some economists believed that banking system is the biggest victim, while there are some economists concluded that banks are in fact the culprit. Trillions of dollars’ worth of housing mortgages defaulted and dodgy derivatives swilling round, worth a lot less than the bankers had previously expected. Many banks went bankrupt. The reason why banks suffered this huge amount of loss is the fact that level of profits and losses goes along with the level of risks a bank takes. Thus banks with high losses probably have high risks.

Is widespread CRT activity one of the reasons that caused the turmoil? Did impacts of different types of CRT activities differ in risk transfers? How were each CRT instruments influenced by bank characteristics and macro economy? Were CRT activities effective for transferring risks in the crisis? What was the impact of CRT activities on U.S commercial banks? And what changed for CRT activities before crisis, during the crisis, and in recession? To answer all these questions, I draw the research question of this study:

How do bank characteristics and macro economy affect different CRT instruments, and in what way is the engagement of those CRT instruments influencing the risk of

U.S. commercial banks before crisis, during crisis and in recession?

To address the research question stated above, this paper studies a group of U.S. commercial banks focusing on their CRT engagement and development from 2001 to 2010. The next section provides background knowledge, test methodologies and research conclusions from previous papers and summaries. The third section is methodology, which illustrates what tests were used and how they were performed on the dataset. In the data section, I explain which banks are in my sample and which data of the banks were collected and organized for quantitative analysis. The empirical results follow, where the test results are presented, compared and analyzed with theoretical support. Lastly, I will conclude the study and discuss limitations.

Nan 13-8-10 3:50 PM

(4)

!

2. Literature Review

The matter of credit risk has drawn much attention of economists due to the enormous damage of the 2007—2009 crisis. Some financial economists were aware of the unbearable high risk of the financial system, and to some extent, they have foreseen a substantial probability of a financial crisis. Unfortunately, they were unable to stop it, or in another view that it is too late to rescue the risky market. Research on credit risk (transfer) is becoming much more popular ever since the presence of this crisis.

CRT with its innovations can ameliorate financial stability by distributing risks among lots of investors. Even though the total amount of risk remains the same in banking system, CRT reduces the risk that individual banks hold through diversification. Demsetz (1999) has proven the hypothesis that banks cannot diversify internally via loan sales to achieve diversification. Large exposures can be passed to investors as smaller diversified exposures. Under certain circumstances, some risks are transferred out of the banking system, for example to institutional investors, hedge funds and equity investors in specialty finance companies.

Besides the above discussion based on the precondition that risks transferred from banks to the system remains at the same amount, other research finds that risks reduced in individual bank along with greater risks in financial system. Nijskens and Wagner (2010) find that, the share price beta of those banks which trade CDS (Credit derivatives swaps) or issue CLO (Collateralized loan obligations) increases significantly. They further the research by separating this beta effect into volatility and a market correlation component. By doing so, the decomposition shows that the increase in the beta is solely due to an increase in banks’ correlations. Hence, CRT causes inter dependence among financial system, thus leading to potential crisis.

Nan 13-8-10 3:52 PM

(5)

Financial economists on credit risk debate about the role of CRT in exacerbating the 2007–2009 crises. Researchers pay high attention to the usage and effects of loan sales, securitizations, and credit derivatives over the last decade. Bedendo and Bruno (2012) studied how bank characteristics and macro economy motivated CRT engagement in the US commercial banks. They discovered the fact that the funds released through CRT are subsequently invested by banks to sustain credit supply by documenting beneficial effects on economy. However, they report higher overall riskiness in banks that engaged intensively in loans sales and securitization, which can be translated into higher default rates during the crisis. What is worthy noted, that the benefits and drawbacks of CRT are much stronger for loan sales and securitization than for credit derivatives.

(6)

!

There is an analysis stating the banks want to improve their stability by using CRT instruments to transfer credit risks, so an efficient transferring ability is very welcome (Wanger and Marsh, 2006). One of the interesting findings is that the allowance for aggregate risks shedding remains risks in more fragile banking sectors, and the CRT from banks to non-bank institutes is more beneﬁcial than those within the banking sector. Therefore, banks maximize benefits by encouraging the use of cross-sectional CRT instruments.

After reading previous literature, I found using CRT instruments may change the credit risk of a bank, and loan sales and securitizations are more efficient in affecting bank risks than credit derivatives (Bedendo and Bruno, 2012). According to Ashraf, Altunbas and Goddard (2007), bank characteristics such as bank size, liquidity and capital, etc. are very important incentives of loan sales, securitizations and credit derivatives. Besides, Bedendo and Bruno (2012) also found that macro economy such as GDP, stock market index, and real estate market are influential on the CRT instruments as well. These previous studies and findings enable estimations of CRT engagement based on bank characteristics and macroeconomics. Thus, to support the research question, hypotheses can be formulated as following:

(1) for incentives of CRT instruments:

H0 – Bank characteristics and macro economy have no influence on CRT

engagement of a bank.

(2) for influence of CRT on bank risks:

H0 – CRT engagement and macro economy have no influence on bank risks.

3. Methodology

Loan sales, loan securitization, and credit derivatives vary from time to time across different banks. Any increase or decrease can be caused by changes from the bank itself and the macroeconomic environment as brought out by previous literature. To

Nan 13-8-10 3:54 PM

(7)

study how a bank’s characteristics influence those activities and how the changes of bank characteristics, macro economy are translated into increase or decrease of CRT activities, I use pooled OLS model to interpret the relationship between these incentives and CRT activities. I started with investigating bank characteristics and macroeconomic parameters as incentives of CRT activities, that how those incentives are associated with engagement in CRT.

3.1 CRT ratios

Loan sales and Loan securitization

Because loan sales and loan securitizations share very similar features in terms of liquidity origination, capital relief, and CRT (Bedendo and Brunella, 2012), I add them up as one variable in the regression analysis. And I regress the change of loan sales and loan securitizations (LSS) over total assets between quarters on bank characteristics and macroeconomic parameters:

!"#!!"#$%!,!= !!+ !!!!!!!∗ !!,!!!+ !!!!!!∗ !!,!+ !!,! (1) where Xi stand for bank characteristics, and Mj sand for macroeconomic parameters. While βi and γi are regression coefficients of the independent variables. The CRT ratio measures the outstanding assets been sold or securitized. As to the independent variables, bank characteristics are specified as loan composition, bank size, the capital ratio, deposit ratio, liquidity asset ratio, z-score, cost of funding, non-interest income ratio, and the non- performing loan ratio. Loan compositions, which covers different types of loans, including 1-4 family mortgage, other mortgage, C&I (Commercial &Industrial) loans, and loans to individuals (which is formally named as consumer loans) over total loans. Bank size and capital adequacy are two constraints of loan supply in a bank. To better study the loan trades, I include bank sizes as an independent variable which is defined by natural logarithm of total asset. For capital ratio, according what Kishan and Opiela (2000) did to analyze the cross-sectional differences in bank financing and lending decisions. Equity capital ratio is computed

Nan 13-8-10 3:53 PM

(8)

!

as total equity capital to total assets. One the two major functions of CRT instruments is creating cash flows. To account for the vital problem of liquidity, the regression uses three characteristics to assess both asset and liability liquidity, including the asset liquidity ratio, cost of funding, and the deposit ratio. Asset liquidity ratio is estimated as the weight of liquid assets. Funding liquidity is measured by two factors, the funding cost and deposit ratio, which are respectively calculated as the interest expense over total liabilities, and total deposit over total liabilities (Bedendo and Bruno, 2012). Besides, I include the non-interest income ratio, which is non-interest income over total assets, to see how much non-interest income can be made through servicing fee and originations of loans or sales of assets, other than the purpose of transferring risks. Another important consideration is about borrowers’ default risk. Non-performing loan ratio is selected regarding risk of default, of which non-performing loans are loans that are in default or close to default. This risk factor describes the default risk of individual debtors causing bank losses.

(9)

Table I Variable list of equation (1)

In this table, there are all variables of the regression Equation (1). The study will perform four regressions on the same set of independent variables which are Xi and Mj. And these four regressions are for dependent variables of LSS/TA t, LSS1-4/TA t, LSSother/TA t, and CDnet/TA t, in which LSS/TA t is on loan sales and securitizations, while CDnet/TA t is on credit derivatives. As for LSS1-4/TA t and LSSother/TA t, they are categorized as sensitivity test for loan sales and securitizations.

Dependent variables

CRT ratio t

(i)

LSS/TA t

LSS1-4: Loan sales and securitization of 1-4 home mortgages LSSother: Loan sales and securitization of other mortgages CDnet/TA t

Credit derivatives net position =

Short credit protection position - Long credit protection position

Independent variables

X t-1 (i) 1_4 t-1 Mortgages securitized by 1-4 residential family

C_I t-1 Commercial and industrial loans CONSUMER t-1 Loans to individuals

OHTER_M t-1 Other mortgages (detailed list in Data) DEP t-1 Deposit ratio = Total deposit / Total liabilities SIZE t-1 Bank size = ln (Total assets)

CAP t-1 Capital ratio = Total equity capital / Total assets.

LIQUID t-1 Asset liquidity ratio = (federal funds sold + securities purchased under agreement to resell + securities held to maturity + securities available for sale) / total assets NON_INT t-1 Non-interest income ratio = Non-interest income /Total assets NONPERF t-1 Non-performing loan ratios = Non-performing loans / Total loans COST_F t-1 Cost of funding = Interest expense / Total liabilities

M t (j) GDP t GDP = ln(GDP)

STOCK t Stock = ln(S&P 500 index) HOUSE t House = ln(house price index)

To make sure that bank characteristics and macroeconomic parameters are not interdependent, I test correlations between every pair of the independent variables.

Nan 13-8-10 3:57 PM

(10)

!

Appendix II shows that all correlations are small, so they have independent importance in determining CRT ratios. Moreover, concerning possible endogeneity arising from the bank characteristics which may have impact from previous CRT activities, I use the lagged value of bank characteristics in regression, as what is done by Bedendo and Bruno (2012) and Altunbas, Gambacorta and Marques-Ibanez (2009) in order to avoid endogeneity bias. Nevertheless, in order to confirm that if there is no endogeneity problem, I generate all the covariance and correlations of residuals and independent variable after every regression:

!"# !!,!!!, ∈!,! = !0!!!"#!!!"# !!,!, ∈!,! = !0 (2)

Sensitivity test

Although loan sales and loan securitizations are pooled together in the regression, there may be distinct driving forces for sales and securitizations of different loan types. Hence, to analyze CRT instruments corresponding to particular types of loans, regressions are to be performed on loan sales and securitizations of 1-4 family mortgage, and on loan sales and securitizations of other loans respectively. And when studying “LSS1-4”, “Other mortgage / Total loans” should be excluded from the independent variables; and when studying “LSSother”, “1-4 family mortgages / Total loans” should be excluded.

Credit derivatives net position

Very similar to the loan sales and loan securitization, credit derivative are driven by bank characteristics and macroeconomic parameters as shown in Table I. Credit derivatives net ratio is regressed on the same set of independent variables, only excluding the “Other mortgage / Total loans”. The results of these regressions will present if bank characteristics and macroeconomic parameters have different impact on credit derivatives trades.

Nan 13-8-10 4:01 PM

(11)

3.2 Bank risks

Banks are highly motivated in using CRT instruments as they can transfer selected credit risks to third parties according to their preferences and needs. To assess whether banks with more involvement in CRT activities default more frequently or more likely to default than others, I study the relationship between CRT engagement and bank risk, with data of banks involved in at least one type of CRT activities.

Concerning internal and external determinants, I regress bank risks on all detailed types of CRT instruments (including the loan sales and securitizations for 1-4 family mortgage ratio, the loan sales and securitization for other loans ratio, the credit derivatives ratio); and a set of bank characteristics namely capital ratio, deposit ratio, liquid asset ratio and bank size (all defined in Table I), also the revenue growth of the bank, and quarterly growth of GDP as well:

Risk!,!= !!+ !!!!!!∗ !!,!!!+ !!∗ !"#. !"#$%ℎ!+ !!∗ !"!"#!+ !!,! (3) where Xi stand for CRT ratios and bank characteristics, while λi, µi and υi are regression coefficients of the independent variables. The measurement of a bank’s risk is not unique. The most common and frequently used risk indicator of a bank’s riskiness is the standard deviation of the return on total assets ( RoA). The bigger the RoA gets, the higher the bank risk is. Another widely used risk indicator is the natural logarithm of the z-score of the bank, which is calculated with RoA and capital ratio (see Table I). A bank becomes insolvent when bank capital is not enough to offset the losses (Boyd and Runkle, 1993); and a higher z-score identifies lower bank risk. Z-score and RoA have given a view of bank situations at profit and loss level. What is to be mentioned, since z-score is calculated with RoA and capital ratio, when performing regression tests with risk measured by z-score, I exclude capital ratio from the independent variable group.

Nan 13-8-10 4:02 PM

(12)

!

Table&II Variable list of equation (3)

In this table, there are all variables of the Equation (3). The study will perform two regressions on the same set of independent variables which are Xi, revenue growth and GDP index. And these two regressions are for different measurements of risks as σRoA and ln(z-score).

Dependent variables Risk t (i)

RoA Standard deviation of RoAa (return on asset). ln(z-score) Z-score = (mean RoA + mean capital ratio) /σRoA

Independent variables

X t-1 (i) LSS1-4/TA t-1 LSS1-4: Loan sales and securitization of 1-4 home mortgages

LSSother/TA t-1 LSSother: Loan sales and securitization of other mortgages CDnet/TA t-1 Credit derivatives net position =

Short credit protection position - Long credit protection position DEP t-1 Deposit ratio = Total deposit / Total liabilities

SIZE t-1 Bank size = ln (Total assets)

CAP t-1 Capital ratio = Total equity capital / Total assets.

LIQUID t-1 Asset liquidity ratio = (federal funds sold + securities purchased under agreement to resell + securities held to maturity + securities available for sale) / total assets GDP t GDP = ln(GDP)

Rev.Growth t Revenue Growth = Total revenue t / Total revenue t-1 -1

a.!RoA!is!return!(income!before!extraordinary!items!&!other!adjustments)!on!total!assets.!Means!and! standard!deviations!are!calculated!over!4!quarters.!

I examine the correlation between every two variables to make sure that there are no interactions between independent variables. Appendix III presents the correlations, which are all small. What should also be mentioned is that bank risks do not start over at the beginning of each financial period, so previous situations are influential on current level risk. As what is done in model (1) of CRT ratio, I use lagged value of bank characteristics to avoid endogeneity, and then calculate covariance and correlations between residuals and independent variables of each regression as equation (2).

Nan 13-8-10 4:05 PM

(13)

3.2 Robustness test

The global financial crisis took a year from burst to its peak in September 2008, when the investment bank Lehman brothers went bankrupt. It broke the notion that some banks are ‘too big to fail’. The sudden change of banking system came at the beginning of crisis, and lasted for the whole period of crisis till the middle of 2009, and then is followed by recovery in later years. For the purpose of researching on CRT activities in the more precise period during crisis, I performed a robustness test with data from 2007.Q3 to 2009.Q2.

To prove that there is no endogeneity problem in none of my models, I present covariance and correlations between residuals and every independent variables of every test in Appendix IV. These covariance and correlations are all very small, close to zero, which indicates no interactions between residuals and any bank characteristics, macroeconomic factors or CRT ratios. In this respect, there is no sign of endogeneity.

4. Data

To investigate the CRT activities on U.S commercial banks of the first decade in the 21st century, I collected data on balance sheets, income statements, and similar resources of U.S. commercial banks. From Bankscope, I selected a list of commercial banks under limitations of domestic location in the U.S. and with assets over $1 billion on the reporting date over 2004.Q1 to 2010.Q4. The quarterly detailed data is collected from the Consolidated Reports of Condition and Income (Call Reports), whose filling is compulsory for all insured commercial banks and trust companies operating in the U.S. (Bedendo and Brunella, 2012).

(14)

!

origination of banks started to boom in 2004 and reached the peak in 2007. They then fell rapidly during the crisis and recession. But I am looking at the long term on effect of CRT activities. Hence, to discover the impact of CRT activities on bank risk, I used raw data from 2001.Q1 to 2010.Q4, which is the period applied for statistical tests. Banks perform quite differently with different sizes. Hereby banks with assets larger than $1 billion but smaller than $20 billion are defined as median-sized banks, and banks with asset larger than $20 billion are large-sized banks. Because of limited involvement of small banks in CRT market, I exclude small-sized banks with assets lower than $1 billion to make the sample. There are 291 banks meeting the asset and geographical requirements, but of those 291 banks only 139 banks have engaged in at least one of the three CRT activities.

It is a set of panel data of bank and timeline. The data is recorded from Call reports, of 139 banks with their detailed data quarterly from 2001.Q1 to 2010.Q4. I used raw data from 2001.Q1 to 2010.Q4. Since means and standard deviations are calculated over every four quarters, the effective data starts from 2002.Q1.

The major cause of the subprime crisis is that bank increased interest on mortgages or lower the loan ratio1_{of houses, so that more and more repayments of residences} defaulted. House prices fell below the value of mortgages, which made those who could afford the repayments stop paying, leading to further drop of the house price. Hence housing mortgages are mostly traded loans and contribute most to the crisis burst. So the research classified loans on balance sheet into two categories: (1) 1-4 mortgages, which are mortgages securitized by family consisting of 1 to 4 residents, and (2) other loans, which are other mortgages (securitized by family consisting of more than 5 residents, by farmland, by nonfarm nonresidential properties, and loans for construction and loan development), consumer loans (loans to individuals, etc.), and commercial and industrial loans.

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

(15)

A loan sale is a conditional trade of cash streams (from a specific loan) between financial individuals or institutions, usually sold by a bank to remove the loan from its balance sheet. According to Bedendo and Brunella(2012), loan sales is measured as the difference between (1) the outstanding principal balance of assets owned by others with services retained by the bank and (2) the outstanding principal balance of assets sold and securitized by the bank. The Economy Watch has defined loan securitization as the process of pooling various types of receivables and using them as collateral to issue securities.2_{Through this process, banks of loan originators reduce} the risk and adjust the financial structures of their own. So the measurement of loan securitization is the outstanding principal balance of assets sold and securitized by banks with servicing retained or with recourse or other seller-provided credit enhancements (Bedendo and Brunella, 2012). In the same way as loan sales, loan securitization is classified into 1-4 residential family mortgage and the other loans. I follow guidelines of calculating loan sales and securitization provided by Bedendo and Brunella (2012) in the empirical research.

Credit derivatives are used much less compared to the two types of activities above in the CRT market, though they do influence bank’s exposure to credit risk. They are contractual agreements between two investors (private or institutional) on financial assets, namely swaps, forward contracts, options, etc. Credit derivative swaps capture the major market of credit derivatives, whose price is driven by the credit risks of investors. Credit derivatives are frequently used to hedge risks. If banks purchase or sell credit derivative on the purpose of hedging, the purchased amount and sold amount would not be balanced. Thus, just as Minton et al.(2009) measured in their researches, the difference of short and long positions of a bank is net credit position (measurement of credit derivatives activity).

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

(16)

!

Table III summarizes the number of CRT instruments-using banks in this study. It is assured that every bank is involved in at least one type of activities. The use of CRT activities has steadily increased along with the timeline. Among all types of CRT, loan sales and loan securitizations are much frequently used than credit derivatives.

Table III Bank engagement of CRT activities from 2001 to 2010 This table presents the number of banks that used different CRT. All loan sales and loan securitizations are added up in the 2nd column, and divided into two groups by types of loans/mortgages in column 3 and 4. The 5th_{column on the right hand side shows credit} derivatives activity in use among the total 139 banks.

Number of CRT activities in use by banks Time Loan sales &

Securitizations

LSS: 1-4

family mortgage LSS: Other loans

Credit Derivatives 2001 101 91 33 13 2002 109 95 45 14 2003 112 97 52 17 2004 108 94 52 16 2005 113 97 54 18 2006 116 98 61 20 2007 119 101 68 22 2008 121 103 68 24 2009 123 107 70 28 2010 123 110 70 27

(17)

Table IV CRT activities and loan compositions in different periods of time The table consists of two parts. The upper part is the list of means of ratios of three CRT activities and weights of each loan component in banks’ financial structure. The lower part is the same information in median values. Means and medians are calculated in two sets of data in different periods: before crisis, and the crisis & recession. Statistical tests are applied to compare if there are differences of loan portfolios or sales/securitizations in two periods.

Call Report Items Before-crisisc Crisis &

Recessiond Statistical tests

Mean t-test Stdev.

Loan sales &

Securitizations Ratio 0.2064 0.2498 -3.0319*** (0.0024) 1-4family mortgage 0.1527 0.1962 -3.4168*** (0.0006) Other loans 0.0536 0.0536 -0.0003*** (0.0098) CDanet Ratio -0.0029 -0.0045 1.4275*** (0.0035) 1-4 family mortgage/TL 0.2357 0.2317 1.8804*** (0.0187) Other mortgage/TL 0.2656 0.2866 -4.4743*** (0.0062) C & Ib loans/TL 0.2193 0.2115 1.9315*** (0.0535) Loans to individuals/TL 0.1201 0.0988 4.2687*** (0.0044)

Median Wilcoxon Stdev.

Loan sales &

Securitizations Ratio 0.0320 0.0497 4.5484*** (0.0103) 1-4 family mortgage 0.0083 0.0172 -3.7522*** (0.0039) Other loans 0.0000 0.0000 7.2202*** (0.0255) CD net Ratio 0.0000 0.0000 2.6291*** (0.0086) 1-4 family mortgage/TL 0.2176 0.2232 -1.6475*** (0.1073) Other mortgage/TL 0.2569 0.2805 -4.5156*** (0.0874) C & I loans/TL 0.1894 0.1852 2.5263*** (0.0115) Loans to individuals/TL 0.0674 0.0395 8.6893*** (0.0063)

a. CD stands for credit derivatives

b. C&I is the abbreviation of Commercial and Industrial c. Before-crisis: 2002. Q1 – 2007. Q2

d. Crisis & Recession: 2007. Q3 – 2010. Q4

e. *** significant at 1% ; ** significant at 5%; * significant at 10%.

The period of before crisis is from 2002.Q1 to 2007.Q2, which has covered the low level period of housing mortgages, and the boom period (2004 to 2006). Though there

Nan 13-8-10 4:06 PM

(18)

!

were booming years, the data shows more involvement in sales and securitization of 1-4 mortgages in crisis than before. However, in later section of regression analysis, I tested data of specific sub periods to see the detailed changes of CRT activities’ effect. From the results of t-tests and Wilcoxon tests in Table IV, it is obvious that in the early years before crisis banks sold and securitized more of the 1-4 mortgages than other loans. Meanwhile, banks grant many types of loans. In this research, I focus on four of those loans which are the major parts of all loans. As listed in Table IV, the four loans are 1-4 family mortgage, other mortgage, C&I (Commercial & Industrial) loans, and loans to individuals (which is formally named as consumer loans). There are significant increase in weights of other mortgages and significant decrease in weights of loans to individuals during the crisis and recession.

5. Empirical Results

Models have been tested with methodology and data discussed above. I summarize results into tables and interpret those results in the following text. This section is divided into two parts: (1) the CRT instruments with their incentives, (2) the bank risks with CRT instruments and other influential parameters.

5.1 CRT ratios

The first hypothesis is about how bank characteristics and macro economy motivate CRT usages. The results are presented in Table V, Table VI and Table VII. These tables include coefficients of every independent variables and corresponding standard deviations. The appropriate regression statistics of adjusted R-square and F-statistic are included as well. Table V demonstrates the effects of incentives on LSS and credit derivatives in two periods: before crisis and during crisis & recession.

Table V Incentives of CRT ratios

The table presents the results of regression tests of equation (1), which is about CRT ratios on their incentives. The CRT ratios in this table are the pooled factor of loan sales and loan securitizations, and the credit derivatives protection positions. They are regressed on bank

Nan 13-8-10 4:10 PM

(19)

characteristics and macroeconomic parameters. Tests use lagged value of#bank#characteristics# # to#avoid#problem#of#endogeneity.

LSS ratio CDnet ratio

Before-crisisa Coefficient Stdev. Coefficient Stdev.

1-4 mortgages t-1 0.7885*** (0.0871) 0.0258*** (0.0056) C&I loans t-1 0.0919*** (0.0971) 0.0133*** (0.0063) Capital ratio t-1 -0.3002*** (0.1779) -0.1999*** (0.0154) Cost of funding t-1 1.9458*** (0.6173) -0.2085*** (0.0555) Deposit ratio t-1 -0.5014*** (0.0615) 0.0138*** (0.0056) Consumer loans t-1 0.4340*** (0.0755) 0.0301*** (0.0047)

Liquid asset ratio t-1 -0.9738*** (0.0729) 0.0111*** (0.0066)

Non-interest income ratio t-1 0.4497*** (0.7074) 0.1708*** (0.0641)

Non-performing loans t-1 0.3037*** (0.8431) -0.0262*** (0.0757) Other mortgages t-1 -0.2028*** (0.0839) Log(TA) t-1 0.0462*** (0.0056) 0.0009*** (0.0005) GDP t 0.4915*** (2.1456) 0.1474*** (0.1949) House t -0.3273*** (0.6132) -0.0329*** (0.0557) Stock t 0.0311*** (0.2305) -0.0128*** (0.0209) Adjusted R-square 0.2152 0.0879 F-statisic 57.7340*** 22.4707***

Crisis and Recessionb Coefficient Stdev. Coefficient Stedv.

1-4 mortgages t-1 1.4657*** (0.1264) 0.0002*** (0.0082) C&I loans t-1 0.4501*** (0.1393) -0.0725*** (0.0089) Capital ratio t-1 -0.1679*** (0.2837) 0.0059*** (0.0240) Cost of funding t-1 1.1037*** (1.1005) -0.6625*** (0.0949) Deposit ratio t-1 -0.4647*** (0.1165) -0.0145*** (0.0100) Consumer loans t-1 0.6975*** (0.1146) -0.0156*** (0.0069)

Non-interest income ratio t-1 -0.7080*** (1.1357) 0.3861*** (0.0976)

Non-performing loans t-1 2.0071*** (0.4575) 0.0038*** (0.0394) Other mortgages t-1 -0.0863*** (0.1152) Log(TA) t-1 0.0464*** (0.0080) 0.0004*** (0.0007) GDP t -0.9118*** (1.1718) 0.1490*** (0.1012) House t 0.1786*** (0.3254) 0.0937*** (0.0278) Stock t 0.0490*** (0.1122) -0.0160*** (0.0097) Adjusted R-square 0.1748 0.0910 F-statisic 30.2161*** 15.8718***

a. Before-crisis is the period of 2002.Q1 – 2007.Q2 b. Crisis and Recession is from 2007.Q3 – 2010.Q4

c. *** significant at 1%; ** significant at 5%; * significant at 10%.

Nan 13-8-10 4:10 PM

![11]:!Change!prob.!Into!stdev.!

Nan 13-8-10 4:11 PM

(20)

!

Bank size, one of the most important determinants, is positively related to both the LSS ratio and CDnet ratio, which means that larger banks tend to sell / securitize more loans and trade more credit derivatives than smaller banks. In column 2-3 of Table V, I observe that the amount of 1-4 family mortgage positively related to amount of loan sales and securitization, which means banks with more weight in 1-4 family mortgage had more of their loans sold and securitized. It is in line with the fact of house and credit boom between 2004 and 2006 (Calem, Henderson, and Liles, 2011). When moving to the period of crisis and recession, 1-4 family mortgage still motivated sales and securitizations of loans. But according to Table IV, LSS has significantly declined in crisis and recession than before. So it can be concluded that banks with loan portfolios weigh more in 1-4 family mortgage sell and securitize more loans regardless of the burst of Subprime crisis. However, other mortgages show negative motivation before crisis and no effect after crisis burst on LSS ratio, which could be biased by mixing all types of loans’ effects on the pooled measurement of sales and securitization of all loans. Hence, in the following results of sensitivity test, I can observe more specific motivation on each type of loans.

(21)

Table VI Sensitivity test on incentives of CRT ratios The table presents the results of equation (1) with dependent variables of LSS1-4/TA and LSSother/TA. This sensitivity test studies loan sales and securitization of 1-4 mortgage and those of other loans respectively on corresponding type of loans. The test uses lagged value ! of!bank!characteristics!to!avoid!problem!of!endogeneity.!

(Over Total assets) LSS1-4 ratio LSSother ratio

Before-crisisa Coefficient Stdev. Coefficient Stdev.

1-4 mortgages t-1 0.9708*** (0.0530) C&I loans t-1 0.2340*** (0.0603) 0.0369*** (0.0317) Capital ratio t-1 -0.2578*** (0.1473) 0.1111*** (0.0825) Cost of funding t-1 0.7346*** (0.5301) 1.4092*** (0.3111) Deposit ratio t-1 -0.2200*** (0.0531) -0.2658*** (0.0310) Consumer loans t-1 0.0003*** (0.0449) 0.5678*** (0.0292)

Liquid asset ratio t-1 -0.9194*** (0.0632) -0.0632*** (0.0368)

Non-performing loans t-1 -0.5682*** (0.7222) 1.2021*** (0.4224) Other mortgages t-1 -0.0120*** (0.0298) Log(TA) t-1 0.0377*** (0.0048) 0.0091*** (0.0028) GDP t 0.4156*** (1.8604) 0.2514*** (1.0862) House t -0.3805*** (0.5318) 0.0142*** (0.3105) Stock t 0.0527*** (0.2000) -0.0257*** (0.1167) Adjusted R-square 0.2043 0.2502 F-statisic 58.2320*** 75.3557***

Crisis and Recessionb Coefficient Stdev. Coefficient Stdev.

Non-interest income ratio t-1 -0.6396*** (1.0173) -0.1240*** (0.4551)

(22)

!

F-statisic 36.3206*** 21.2784***

a. Before-crisis is the period of 2002.Q1 – 2007.Q2 b. Crisis and Recession is from 2007.Q3 – 2010.Q4

c. *** significant at 1%; ** significant at 5%; * significant at 10%.

To find out different key drivers for different types of loans, I looked at the test results of the sensitivity test that regressed LSS1-4 ratio on 1-4 family mortgages and LSSother ratio on other mortgages respectively. Table VI shows that more 1-4 family mortgages lead to more sales and securitization of itself along the first decade of 2000s, while other mortgages had no impact on sales and securitizations of other loans during the same years. This illustrates that 1-4 family mortgage is the key driver to CRT instrument of loan sales and securitization for transferring bank’s credit risks

to third parties. It is argued that banks preserve more capital for a capital shock with

more liquid assets (Altunbas, Gambacorta and Marques-Ibanez, 2009). And in column 2-3 I find negative effect of liquid asset ratio on LSS1-4 ratio, which confirms that banks sell and securitize loans to balance their liquidity as well as transferring borrowers’ default risks. For the same reason, capital ratio should play a similar role as liquidity. However, the test result for other loans says the opposite that higher capital ratio results in more sales and securitization. Note worthily, non-interest income ratio plays an important role for both types of LSS ratios negatively. It is due to the decrease of loans’ origination and distribution by bank in crisis and recession to adapt to the balance of those credit risks which landed back to banks after transferred

to market (Bedendo and Brunella, 2012).

The sensitivity test divides data panel by different types of loans. However, I also looked into the data at the time horizon. So I perform a robustness test on the certain data groups by dividing timeline into before-crisis, crisis and recession, in which the focus is on crisis (from 2007.Q3 to 2009.Q2). Results are presented in Table VII.

(23)

from 2007.Q3 to 2009.Q2. The tests use lagged value of#bank#characteristics#to#avoid# problem#of#endogeneity.

(Over Total assets) LSS ratio CDnet ratio

Crisis Coefficient Stdev. Coefficient Stdev.

1-4 mortgages t-1 1.0741*** (0.1507) 0.0069*** (0.0129) C&I loans t-1 0.1524*** (0.1657) -0.0728*** (0.0138) Capital ratio t-1 -0.5803*** (0.3411) 0.0224*** (0.0366) Cost of funding t-1 2.1469*** (1.4793) -0.7870*** (0.1598) Deposit ratio t-1 -0.5135*** (0.1410) -0.0200*** (0.0153) Consumer loans t-1 0.7594*** (0.1355) -0.0026*** (0.0109)

Non-performing loans t-1 1.8990*** (0.7475) 0.0467*** (0.0812) Other mortgages t-1 -0.2383*** (0.1409) Log(TA) t-1 0.0578*** (0.0101) 0.0006*** (0.0011) GDP t -0.4233*** (1.6282) -0.0818*** (0.1769) House t -1.1035*** (1.3554) 0.4631*** (0.1473) Stock t 0.2606*** (0.2353) -0.0745*** (0.0256) Adjusted R-square 0.1939 0.0798 F-statisic 19.9514*** 8.3623***

LSS1-4 ratio LSSother ratio

Crisis Coefficient Stdev. Coefficient Stdev.

Non-interest income ratio t-1 -0.9802*** (1.2098) -0.1257*** (0.6759)

Non-performing loans t-1 1.8069*** (0.6475) 0.1760*** (0.3608) Other mortgages t-1 -0.0251*** (0.0535) Log(TA) t-1 0.0505*** (0.0087) 0.0080*** (0.0049) GDP t -0.6012*** (1.4115) 0.3049*** (0.7874) House t 0.0402*** (1.1749) -1.2565*** (0.6553) Stock t 0.0697*** (0.2040) 0.2093*** (0.1138) Adjusted R-square 0.1965 0.1945 F-statisic 20.2317*** 21.4869***

(24)

!

By comparing coefficients of same incentives in different periods, I found that bank characteristics and macroeconomic parameters have similar, and almost every one of the significant incentives have larger, influences on CRT ratios in crisis than they have for the period of crisis with recession. Thus, the recession can be viewed as a buffer for all changes in the crisis, that banks changed substantially when crisis burst and bounced back somewhat in recession. The liquidity problem is the primary cause of the crisis and lead to countless mortgage defaults in the crisis.

Apart from bank characteristics, I surprisingly found that macro economy has no influence on CRT activities before the crisis and bare impact during subprime crisis, and the main impact was on trades of credit derivatives. Results show that higher house prices leads to more sale of credit derivatives while good stock market increases the purchase of credit derivatives. It is probably because expensive houses make people borrow more mortgages from banks and banks then transfer those risks by selling more credit derivatives.

5.2 Bank risks

The second hypothesis is about how CRT ratios, bank characteristics and macro economy affect bank risks, and the results are presented in Table VIII. The table includes coefficients of every independent variables and corresponding standard deviations, and appropriate regression statistics of adjusted R-square and F-statistic.

Table VIII Impact of CRT ratios and other parameters on Bank Risks The table presents results of bank risks motivated by CRT ratios, bank characteristics, and macroeconomic parameters. The regressions are on different periods: before-crisis, crisis & recession, and recession only. Risks are measured in two ways and the tests use lagged value of CRT ratios and bank characteristics to avoid problem of endogeneity.

ln(z-score) σRoA

Before-crisisa Coefficient Stdev. Coefficient Stdev.

LSS: 1-4 t-1 -0.1156*** (0.0252) 0.0000*** (0.0001)

LSS: Other loans t-1 -0.1926*** (0.0424) 0.0020*** (0.0002)

(25)

Log(TA) t-1 0.0164*** (0.0069) -0.0001*** (0.0000)

Liquid asset ratio t-1 -0.0665*** (0.0828) -0.0008*** (0.0004)

Deposit ratio t-1 0.5018*** (0.0769) -0.0019*** (0.0003) Capital ratio t-1 0.0106*** (0.0009) Revenue growth t -0.0060*** (0.0123) 0.0000*** (0.0001) GDP t 0.3354*** (0.2189) -0.0019*** (0.0009) Adjusted R-square 0.0359 0.1215 F-statisic 14.4938*** 45.5023***

Crisis & Recessionb Coeff. Stdev. Coeff. Stdev.

LSS: 1-4 t-1 -0.0801*** (0.0446) -0.0006*** (0.0003)

LSS: Other loans t-1 -0.0064*** (0.0961) 0.0012*** (0.0007)

CD net ratio t-1 -0.4415*** (0.4655) 0.0051*** (0.0031)

Log(TA) t-1 0.0676*** (0.0148) -0.0003*** (0.0001)

Liquid asset ratio t-1 0.5531*** (0.2274) -0.0102*** (0.0016)

Deposit ratio t-1 0.1625*** (0.2135) -0.0012*** (0.0014) Capital ratio t-1 -0.0048*** (0.0034) Revenue growth t -0.0254*** (0.0185) 0.0001*** (0.0001) GDP t 2.8668**** (1.4190) -0.0253*** (0.0095) Adjusted R-square 0.0156 0.0369 F-statisic 4.8092*** 8,1799***

Crisisc Coeff. Stdev. Coeff. Stdev.

LSS: 1-4 t-1 -0.0642*** (0.0496) -0.0010*** (0.0004)

LSS: Other loans t-1 -0.0824*** (0.0831) 0.0015*** (0.0007)

CD net ratio t-1 -0.0345*** (0.4059) 0.0032*** (0.0032)

Log(TA) t-1 0.0549*** (0.0152) -0.0003*** (0.0001)

Liquid asset ratio t-1 0.4098*** (0.2356) -0.0116*** (0.0020)

Deposit ratio t-1 0.1308*** (0.2138) -0.0051*** (0.0017) Capital ratio t-1 0.0001*** (0.0039) Revenue growth t -0.0862*** (0.0326) 0.0003*** (0.0003) GDP t 1.2826*** (1.2799) -0.0289*** (0.0102) Adjusted R-square 0.0170 0.0530 F-statisic 3.3859*** 7.8566***

a. Before-crisis is the period of 2002.Q1 – 2007.Q2 b. Crisis and Recession is from 2007.Q3 – 2010.Q4 c. Crisis is the period of 2007.Q3 – 2009.Q2

(26)

!

The two measurements of bank risks are z-score and σ(RoA). They describe risks of the probability of banks with insufficient capital or incapable when confronted with liquidity or capital shocks. When risk is measured by z-score (a higher z-score means a lower risk), I observed the fact that more usage of LSS activities increases bank risks in the years before crisis. But this effect seems against the risk transferring ability of CRT. This finding suggests that sample banks report higher risks with more engagement in sales and securitization of loans. Nevertheless, CRT instruments also function as funding tools, which generate cash flows with sales, securitizations and credit derivatives trade. So the CRT activities in this study perform more of the funding role rather than the risk transferor. Similar regression is performed when bank risk is measured by σ(RoA). Loan sales and securitizations are positively associated with the volatility of return on bank’s total asset before crisis. But according to Bedendo and Bruno (2012) these effects will only be seen among median-sized not large-sized banks, which is due to the fact that active LSS users generates low z-score mostly because of high leverage ratios instead of high volatility in asset returns, as they benefit from the stabilizing effect of significantly influential non-interest income.

Apart from sales and securitization of loans, I also observed another CRT instrument, the net credit protection ratio. However, the results do not suggest a close relationship between CD net ratio and bank risks in any period of time. There is a significantly negative impact on z-score in the years before crisis, indicating higher probability of default. But I cannot see changes of σ(RoA) caused by usage of CD net ratio. The loose association between CD net ratio and bank risks can be explained by Table III of CRT distribution of sample banks that most medium sized banks focus on loan

sales and securitizations. While Bedendo and Bruno (2012) found that large banks

purchase rather than sell credit protection via credit derivatives are less risky.

(27)

This paper investigated bank characteristics and macroeconomic parameters as incentives of CRT instruments, studying the way they motivate CRT instruments, and how bank risks changed by using these CRT activities in different time periods (before-crisis, crisis and recession). Loan sales and loan securitizations, which are two major CRT instruments, have been pooled together as one independent variable in the test, and another major CRT instrument is net Credit Derivatives Protection. According to sample bank data, loan sales and securitizations are much more frequently used than credit derivatives trades. The research is about U.S. commercial banks in the first decade of the 21st century. I select banks sizes under constraints on total assets measuring certain bank size.

Answers to the research questions are found in the test results, which suggest that bank characteristics and macroeconomic parameters exert more important influence in crisis and recessions for loan sales and securitizations (LSS). The LSS ratio goes for different types of loans, 1-4 family mortgage which is mortgage securitized of 1-4 residential family and the other loans including other mortgages, commercial & industrial loans, and consumer loans. The findings show banks that increase their holding of 1-4 family mortgages tend to sell or securitize more loans, while the holding of other loans does not affect the usage of this CRT instrument, no matter before or after crisis. And banks of larger sizes tend to be more active in LSS of 1-4 family mortgages, having lower non-performing loan ratios. In this respect, larger banks gather more capital and liquidity without a corresponding increase in risks. Liquid asset ratio closely relates to the engagement of CRT activities of a bank, since banks requires sufficient liquidity to survive large amount of default. So LSS becomes active when liquidity assets weigh less on the banks’ balance sheet.

But credit derivatives are not as sensitive as LSS, neither is the impact in all periods of time as significant as loan activities. From before-crisis to crisis & recession, the

tests results indicate that banks tend to become net sellers and sell more credit

derivatives. It is explained by the fact that short positions in credit derivatives are

Nan 13-8-10 4:17 PM

(28)

!

mostly of single-name credit default swaps which act like an insurance contract against the default of a reference entity (Juurikkala, 2011). By shorting more credit derivatives, banks fund and preserve capitals in case of capital shock. In the meanwhile, house prices and stock market have no influence on LSS and bare impact on credit derivatives.

Concerning the influence of CRT instruments on bank risks, the study suggests that CRT instruments as short-term funding tools have increased return on assets but exposed more risks at the same time. In this study, banks with more engagement in CRT ratios have higher risk to default in the economic downturn. The potential inference is that the more risks those banks transferred, the more risks that could have landed back at themselves. CRT played the role of a funding tool more than a risk transferor. The higher level of risks may lead to the subprime crisis. And similarly, credit derivatives trades had barely any influence on bank risks while loan sales and securitization stimulated the risk level of banks.

This quantitative research is limited in several ways. Firstly, the sample is limited to banks which have $1 to $20 billion in total assets. These are banks at medium size, so the test results may not be valid for banks of larger size. Secondly, some other independent variables cannot be included in the regression due to the unavailability of some data. Like Bank Holding Company (BHC), the regulation permits an insurer to turn expected losses regarding the failure of a banking subsidiary to the capital of affiliated banks which do not fail (Ashcraft, 2008). The two common categories are single and multi-BHC, which are influential factors on bank risks. Being unable to define the BHC category of all banks in the sample, so I did not include the factor in the regression. It is the same situation for loan ratio, which is one minus the down payment over total price of the house. Thirdly, the Before-Crisis period is tested from 2002.Q1 to 2007.Q2, but real estate loan sales and CRT instruments usage increased dramatically from 2004. So transfer ability of the overall period may be underestimated. Lastly, the limitation is due to the scope of my research. Even if it

Nan 13-8-10 4:20 PM

(29)

takes quite a few factors as drivers of CRT instruments, I believe there are probable some factors I am not familiar with and may be omitted. This could bias the influence of the current factors or even the results’ interpretation.

7. Reference

Altunbas, Y., Gambacorta, L. and Marques-Ibanez, D. 2009. Securitization and the bank lending channel. European Economic Review, 53, 996-1009.

Ashcraft, A.B., 2008. Are Bank Holding Companies a Source of Strength to Their Banking Subsidiaries? Journal of Money, Credit and Banking, 40, 2–3.

Ashraf, D., Altunbas, Y. and Goddard, J., 2007. Who Transfers Credit Risk? Determinants of the Use of Credit Derivatives by Large U.S. Banks. The European Journal of Finance, 13, 483-500.

Bedendo, M. and Bruno, B. 2012, Credit risk transfer in U.S. commercial banks: What changed during the 2007–2009 crisis? Journal of Banking & Finance, 36, 3260–3273

Boyd, J.H., Runkle, D.E., 1993, Size and performance of banking firms. Journal of Monetary Economics 31, 47–67.

Buffett, W., 2002. Chairman’s letter. Berkshire Hathaway 2002 Annual Report. Calem, P., Henderson, C., and Liles, J., 2011. “Cherry picking” in subprime mortgage securitizations: Which subprime mortgage loans were sold by depository institutions prior to the crisis of 2007? Journal of Housing Economic, 20, 120-140.

Chies, G. 2008, Optimal credit risk transfer, monitored finance, and banks. J.Finan. Intermediation, 17, 464–477.

Demsetz, R.S. 1999. Bank Loan Sales: A New Look at the Motivations for Secondary Market Activity. Journal of Financial Research. 23, 197-222.

Greenspan, A., 2005. Economic flexibility. Speech to the National Association for Business Economics Annual Meeting. Chicago, September 27, 2005.

(30)

!

Minton, B.A., Stulz, R.M., and Williamson, R.G., 2005. How much do banks use credit derivatives to reduce risk? NBER WP N. 11579.

Nijskens and Wagner 2010, Credit risk transfer activities and systemic risk: How banks became less risky individually but posed greater risks to the financial system at the same time. Journal of Banking & Finance, 35, 1391–1398

Parlour, C.A., Winton, A., 2013. Laying off credit risk: loan sales versus credit default swaps. Journal of Financial Economics, 107, 25-45.

(31)

Appendix I

Table&IX&Data&items&recorded&from&Call&Report

Data collection

Total loans RCFD1400 (gross total loans and leases)

Total assets RCFD2170

1-4 mortgages RCON1430

Other mortgages RCON1415 + RCON1420 + RCON1460 + RCON1480

Commercial and industrial loans

RCFD1600 (C&I loans) + RCFD1590 (agricultural loans)

Consumer loans RCFD1975

Non-performing loans RCFD1407 (loans over 90 days late) + RCFD1403 (non-accruing loans)

Total deposits RCFD2200

Liquid assets RCFD1350 (fed funds sold and securities purchased under

agreement to resell)

+RCFD1754 (securities held to maturity) +RCFD1773 (securities available for sale) Return-on-Assets

(RoA)

RIAD4300 (income before extraordinary items and other adjustments) / RCFD2170(total assets)

Interest expenses RIAD4073

Total liabilities RCFD2948

Total capital RCFD3210

Non-interest income RIAD4074

z-Score (Mean RoA + mean capital ratio)/standard deviation of RoA.

Mean RoA, mean capital ratio and standard deviation of RoA are computed over the past four quarters

Total revenues RIAD4107 (interest income)

+ RIAD4074 (non-interest income) Securitization of 1-4 mortgages RCFDB705 Securitization of other loans RCFDB706 + RCFDB707 + RCFDB708 +RCFDB709 + RCFDB710 + RCFDB711 Sales of 1-4 mortgages RCFDB804 + RCFDB805 - securitization of 1-4 mortgages

Sales of other loans RCFDA591 - securitization of other loans

Credit protection purchased via CD

For period :2001.Q2-2005.Q4 RCFDA535

For period 2006.Q1-2010.Q4 RCFDC969 + RCFDC971 + RCFDC973 + RCFDC975

Credit protection sold via CD

For period 2001.Q2-2005.Q4 RCFDA534

(32)

Appendix II

! !

Table X Correlations between independent variables of CRT ratio

The table shows correlations between two two incentives of CRT, including bank characteristics and macroeconomic parameters. Correlation 1_4 t-1 C_I t-1 CAP t-1

COST_F t-1 DEP t-1 CONSU MER t-1 LIQUID t-1 NON_IN T t-1 NONPERF t-1 OHTER_ M t-1

(33)

Appendix III

Table XI Correlation between independent variables of bank risk test

There are the correlations of 9 independent variable of the regression on bank risk in the table, including bank characteristics and macroeconomic parameters.

Correlation LSS1-4/TA LSSother/TA CDnet/TA SIZE t-1 LIQUID t-1 DEP t-1 CAP t-1 Rev.Growth t GDP t

(34)

Appendix IV

!

Table XII Covariance and correlation between independent variables and residuals of regressions of model (1)

There are covariance and correlations between residuals and independent variables of every regression on different CRT ratio and different time periods. A stands for years before crisis from 2001.Q1 to 2007.Q2; B is for crisis of 2007.Q3 to 2009.Q2, and C covers the crisis & recession period 2007.Q3-2010.Q4. All covariance and correlations are very close to 0 when the number is effective with 4 digits decimal. Thus there is no indication of endogeneity.

Residual Cov T-1 T

Cor

1_4 C_I CAP COST_F DEP INDIV LIQUID NON_INT NONPERF OTHER_M SIZE GDP HOUSE X

(35)

(36)

Table XIII Covariance and correlation between independent variables and residuals of regressions of model (3) There are covariance and correlations between residuals and independent variables of every regression on different risk measurements and different time periods. A stands for years before crisis from 2001.Q1 to 2007.Q2; B is for crisis of 2007.Q3 to 2009.Q2, and C covers the crisis & recession period 2007.Q3-2010.Q4. All covariance and correlations are very close to 0 when the number is effective with 4 digits decimal. Thus there is no indication of endogeneity.

Residual Cov T-1 T

Cor

LSS1-4/TA LSSother/TA CDnet/TA SIZE t-1 LIQUID t-1 DEP t-1 CAP t-1 Rev.Growth t GDP t