The effect of securitization on a bank’s credit risk before, during
and after the financial crisis
Name: Thierry Schilder Student number: 11416785
Thesis supervisor: Dr. A. Sikalidis Date: June 25, 2018
Word count: 13,357
MSc Accountancy & Control, specialization Control Amsterdam Business School
2 Statement of Originality
This document is written by student Thierry Schilder who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
3
Abstract
This study investigates the effect of securitization before, during and after the financial crisis on a bank’s credit risk. The data of US bank holding company is used from 2003 to 2017 divided in three periods. A bank is a securitizing bank when it is involved in securitization in one of the quarters in the measured period. Credit risk is measured with proxies risk-weighted assets and charge-off ratio. I find a positive effect of securitization on credit risk for the period before the financial crisis. During the financial crisis I find a negative relation with relative risk-weighted assets and a positive relation with charge-off ratio. There is no evidence found for a relation after the financial crisis, although the results suggest a positive relation. Therefore, securitization had a positive effect on credit risk before the financial crisis, a less strong effect during the financial crisis, and there is no evidence found for a relation after the financial crisis.
4 Contents
1. Introduction ... 6
2. Literature Review ... 8
2.1 Securitization and the process ... 8
2.2 Securitization and the financial crisis ... 10
2.3 Literature Survey ... 11 3. Hypothesis Development ... 13 4. Methodology ... 14 4.1 Sample ... 14 4.2 Dependent variables ... 15 4.2.1 RWATA ... 15 4.2.2. Charge-off Ratio ... 16 4.3 Independent variables ... 17 5. Empirical Design ... 20
6. Data and Descriptive Statistics ... 21
6.1 Descriptive Statistics ... 21
6.1.1. Descriptive Statistics before financial crisis ... 21
6.1.2. Descriptive Statistics during financial crisis ... 23
6.1.3. Descriptive Statistics after financial crisis ... 25
6.2 Pearson Correlation Matrix ... 26
7. Empirical Findings ... 28
7.1 Before the financial crisis ... 28
7.2 During the financial crisis ... 30
7.3 After the financial crisis ... 32
7.2 Robustness Test ... 34
8. General Discussion ... 36
5
8.2 Limitation and Further Research... 38
8.3 Implications ... 38
9. References ... 39
6
1.
Introduction
From the beginning of this century up to the financial crisis, banks became in an increasing competitive environment. To compete with competitors they had to take more risks and securitization became an effective tool for this, with transforming illiquid assets into liquid assets that could be invested in more profitable investment opportunities (Casu et al., 2011). The increasing use of securitization up to the financial crisis resulted in that securitization had a big impact on in the emergence of the financial crisis (Acharya and Richardson, 2009, p. 195-210).
Prior research is mainly about the effect of securitization on general bank risk and is not related to the financial crisis. In general, previous studies suggest a positive relation between securitization and different kinds of bank risk. Franke and Krahnen (2007, pp. 603-635) concluded that banks that announce to involve in securitization will have a higher stock beta. Baur and Joossens (2006) have done research on the relation between securitization and a bank’s systematic risk and found a positive effect. Dionne and Harchaoui (2003) used a sample over the 1988-1998 period to investigate the effect of securitization on the credit risk of banks and they also found a positive effect. Casu et al. (2011) suggest that securitization results in higher bank risk, that it increases the expansion of credit and results in a bank that holds riskier assets. However, Jiangli and Pritsker (2008) state that securitization can have a negative relationship with bank risk, because securitization is a credit derivative and it therefore decreases insolvency risk.
According to Gebhard (2016, p.169) credit risk has been identified as one of the major causes of the financial crisis that started in 2008 and Allen and Powell (2011) state that the financial crisis highlighted how important and necessary it is to measure credit risk well. Casu et al. (2011, pp. 769-789) concluded that before the financial crisis, banks that were involved in securitization choose assets portfolios with lower credit risk. By generating cash with transforming illiquid assets into liquid assets, securitization can increase the expansion of credit and also increase a bank holding riskier assets (Niu and Richardson, 2006; Chen et al., 2008). However, little research is done about the effect of securitization on credit risk of banks and especially how this effect is developed with a view to the financial crisis.
Since banks are under higher pressure to make profits, cash generated from securitization can be tempting to invest in riskier assets with higher returns (Purnanandam, 2009). In most cases securitization at a bank means that the assets are placed at off-balanced-sheets and are still held by the bank and the bank runs the risk (Vermilyea et al., 2008; Risk
7 Management Credit Card Securitization Manual, 2007). Because securitization has automatically effect on credit risk, I will follow Gebhard (2016, p.169), and Acharya and Richardson (2009, p. 195-210) who relate credit risk and securitization as the main causes of the financial crisis. Therefore I hypothesize that securitization has a positive effect on credit risk before the start of the financial crisis. Because of the huge shock and big consequences caused by the financial crisis in 2008 (Atkinson et al., 2013) and the increasing importance of credit risk measurement (Allen and Powell, 2011), I hypothesize that banks during the financial crisis have lower credit risk compared to before the crisis. Based on Berglund and Mäkinen (2016) who state that when a crisis occurs, banks learn from their mistakes and they regulate their business models, I hypothesize that after the crisis banks have lower credit risk compared to before the crisis.
The aim of this research is to investigate the effect of securitization is on credit risk and the development of this effect in three different timeframes; before the financial crisis, during the financial crisis and after the financial crisis. Credit risk will be measured with the ratio between risk-weighted assets and total assets, and with charge-off ratio. This research tries to contribute to literature by looking at the effects of securitization on risk-taking and asset quality of banks. According to prior studies securitization and credit risk have had a major impact on the emergence of the financial crisis. To prevent from happening again it is important to understand what causes a bank being at risk, what the development of these risks is in combination with the financial crisis and the relationship between securitization and credit risk now. Understanding these aspects is important for economic stability and the role of banks in its environment. Financial institutions like banks have an increasing impact on the economic and therefore it is important to understand what the later effect of securitization is and to prevent another crisis.
The sample that is used to investigate the effect of securitization on credit risk consists of US Bank Holding Companies that are required to report their financial reports in Y-9C reports quarterly. The period before the financial crisis is represented by the first quarter of 2003 up to and including the second quarter of 2008 (Foster and Magdoff, 2009, p.11). The period during the financial is beginning at the third quarter of 2008 and ends after the fourth quarter of 2013 (Reinhart and Rogoff, 2014, p. 50-55). The first quarter of 2014 up to the last quarter of 2017 represents the period after the financial crisis.
This introduction will be followed by extensive literature review. Thereafter the hypothesis development will be discussed followed by the methodology, the empirical design and the
8 data and descriptive statistics. The seventh chapter will present the empirical results. Subsequently, the general discussion will be discussed in the last chapter.
2.
Literature Review
The second chapter examines the literature on securitization The process of securitization and the influence of securitization on the financial crisis will be discussed. This is followed by a literature survey about the influence of securitization on a bank’s general risk.
2.1 Securitization and the process
Securitization is a way in which illiquid assets can be transformed into liquid assets. Specifically, it consists simply said of purchasing loans or mortgages, putting them into a pool, and selling the pool as a security (Joosen and ‘t Westeinde, 2002).
A bank uses securitization to transform its illiquid financial assets into securities that are marketable and tradable. These illiquid assets are for example the right of receiving payments from a customer on a loan and these are traditionally held until maturity (Gorton and Metrick, 2012). In the period before the financial crisis, banks used securitization mainly to transform mortgages into cash. A financial asset like the right of receiving payments from a customer on a loan can be an example of an asset that a bank uses to securitize. A bank pools long-term assets that are quite the same together and transforms this into securities that are interest bearing. A bank can sell these securities to investors and the purchaser of these securities gets the right of the payments on these loans. In this way, a bank transforms illiquid assets into liquid assets that are tradable, and sells it for cash (Aluise, 2017).
A more detailed look tells us that the pool existing of assets is owned by the bank, which generally sells the pool to an SPV (Special Purpose Vehicle). An SPV is an entity that is created by the bank. The only purpose of an SPV is the transactions for which it is created. A SPV has no employees, has not a physical location, makes no substantive economic decisions and it cannot go bankrupt (Gorton and Souleles, 2005). The cash flows sold to the SPV are originated by a bank and these securities are called asset-backed securities (ABS). These ABS are rated and sold in the capital market.
Figure 1 gives an overview of the securitization process through a SPV. The originating firm is visible on top in the figure, offers loans, and actively searches for lending opportunities. To fund these loans, the originating firm can do this by intermediary borrowing
9 or by selling it to the SPV. In figure 1 is the SPV called the Master Owner Trust Pool of Assets. The SPV purchases the cash flows of the loan by selling the asset-backed securities to investors in capital markets. The cash flow rights that are generated by the SPV can be tranched. In a tranched issuance, the principal amount and interest payments are not split evenly, but based on seniority. This can be seen in the last step of figure 1.
When the cash flows are based on seniority, the cash flow distribution is as a waterfall. For example, the issuance exists of a senior tranche, next tranche and a last tranche. Any cash that flows into the SPV, will be first distributed to the senior tranche, then the next tranche and so on. When the underlying assets will not be repaid because of for example defaults, the first losses will be on the account of the last tranche (mostly junior tranches). These settlements make sure that the more senior tranches are less likely to get losses and are therefore higher rated. This results in that the more senior tranches are better rated than the average mortgages in the asset pool that is purchased by the SPV. This method of issuing tranches makes it possible to split up credit risk and put it at the clients that will take it (Gorton and Souleles, 2005).
Figure 1 There are several other reasons why banks and other creators are willing to involve in
securitization to take advantage. The first reason is that the result of securitization is that they can transfer some of the risk to others. This is for example transferring the default risk that a customer will not pay back his loan to investors who are willing to take this risk. The second
10 way banks want to take advantage of this method is that banks can better handle their possible asset-liability mismatches by managing their portfolio and transferring assets from their balance sheet off their balance sheet. The third reason is that securitization has a positive effect on the return on capital, because assets are transferred from the balance sheet to off-balance-sheet. The last way banks take advantage of securitization is that because assets are transferred to off-balance-sheets, banks can increase their liquidity to increase their lending operations. This indicates that the amount of dollars a bank sells as a security they can lent out this amount to new borrowers (Aluise, 2017).
According to Culp (2002, pp. 294-304) securitization is a technique in finance in which not easily transferable assets form trading or financial transactions are transformed into transferable debt securities. She states that securitization is a process of unbundling cash flows from assets and putting them into securities that are available for trading. Jiangli and Pritsker (2008) state that securitization is a credit-derivative that move the risk profile of the asset side of the balance sheet. According to Jiangli et al. (2007) banks that securitize are in general larger than banks that are not involved in securitization. This because of the fixed costs that are involved and larger banks can better deal with this.
2.2 Securitization and the financial crisis
In the period before the start of the financial crisis after the summer of 2008 (Foster and Magdoff, 2009, p.11), a lot different type of securities arose and the securities underwent a financial innovation. The pools that were created by banks and sold as securities generated payments that became very complicated and vague packages. These packages are the Collateralized Debt Obligations, better known as CDO’s. Because of the financial innovation and different types of securities, banks had a lot more options in trading these securities. Historically when a bank bought a mortgage, it would hold the mortgage and received the payments until the end of the lifetime. With all the new options in trading the securities, banks did not intend to hold new loans, but they wanted to securitize and sell them as soon as possible. When a bank has the intention to hold a loan for the whole lifetime, the bank will ensure that the borrower can pay the loan and interests as agreed. When a bank has no intention to hold the loans and want to put them into CDO’s, and then transfer all the risk to investors, the capability of the borrower to make the agreed payments become less important (Kolb, 2010, pp. 21-22).
11 Acharya and Richardson (2009, p. 195-210) state that securitization played a big role in the financial crisis that started in 2008. They argue that there were two ways in which bank tried to dodge regulations to perform securitization. They state that this is the main reason of the recent financial crisis. The first way bank tried to avoid the regulations was that they transferred their assets into entities that were off-balance. These were in particular securitized mortgages of large size. Because they placed these assets off-balance, they did not have to keep buffers for them. This was a way to avoid the regulations about holding buffers against mortgages. The second way banks tried to avoid the regulations was based on the rule that banks were allowed to hold a lower amount of capital against their assets if the rating of the assets were AAA. When banks changed the mortgaged in securitized mortgages, they were not limited by the regulations to hold that much assets against their credits. In this way, they could increase their ability to provide many more loans. This situation occurred because of the extreme growth of securitization. The result of this was that almost all the risk of banks focused on the risk related to mortgages, especially mortgage defaults. These assets became worthless when the housing bubble popped.
2.3 Literature Survey
Several studies investigated the relation between securitization and bank risk and concluded that the relation is positive. Franke and Krahnen (2007, pp. 603-635) and Baur and Joossens (2006) investigated this relation to look at the issuance of CDO’s. Franke and Krahnen studied the relation between the stock beta of a bank and the issuance of CDO’s around the announcement date. They conclude that the stock beta will be higher, when a bank announces to involve in securitization. They also state that the risk of banks will be higher when they invest the liquid assets that are generated from securitization into other securitized loans. Baur and Joossens investigated the effects on systematic risk of banks as a result of securitization by looking at the issuance of CDO’s. Their conclusion is that there is a positive correlation between the issuance of CDO’s and the banking sectors excessive joint movements in equity returns. They agree with Franke and Krahnen that it depends on what a bank does with the generated cash what the effect is on the systematic risk of the bank. Also the conclusion of Casu et al (2011, pp. 769-789) is somewhere the same. They also state that there is a positive relation between securitization and a higher risk exposure and that it depends on their investments after securitization activities. Nijskens and Wagner (2011, pp. 1391-1399) investigated the relation between credit risk activities and bank risk by looking at the share price before the crisis and focusing on the systematic risk of banks. They find a higher
12 systematic risk in combination with credit risk activities. Dionne and Harchaoui (2003) conclude that banks have higher credit risk when they are involved in securitization activities. Haensel and Krahnen (2007) and Franke and Krahnen (2005) concluded that issuing CDO’s results in higher systematic risk of banks that issue the CDO.
Although several studies found a positive relationship between securitization and bank risk, several studies concluded that securitization and bank risk are negatively related. As mentioned before, Casu et al (2011, pp. 769-789) state that the relation between bank risk and securitization is positive, but they also conclude that there is a negative effect between securitization and the credit risk-taking of banks. This indicates that banks that involve in securitization with higher amounts, are more risk-averse and therefore being at less risk. Jiangli and Pritsker (2008) also state that securitization can have a negative relationship with bank risk. They argue that since securitization is a kind of credit-derivative that moves the risk profile of the asset side of the balance sheet of the bank and it decreases insolvency risk of a bank. This outcome matches with the conclusion of the investigation of Ben Salah and Fedhila (2012). They compared securitization to bank stability and risk-taking. They state that securitization decreases bank risk and has a positive effect on bank stability.
13
3.
Hypothesis Development
In the previous sections the literature is discussed about the relation between securitization and bank risk. This part of the chapter will discuss the development of the hypotheses. The aim of this research is to investigate the effect of securitization on a bank’s credit risk in the period before, during and after the financial crisis.
There are several studies that conclude a positive relation between securitization and bank risk. Franke and Krahnen (2007), Baur and Joossens (2006) and Dionne and Harchaoui (2003) state that securitization has a positive relation with stock beta, systematic risk and a bank’s credit risk respectively. According to Chen et al. (2008) and Nui and Richardson (2006) it depends on the decision of a bank what they do with the generated cash from the securitization activities what the exposure to credit risk is. According to Acharya and Richardson (2009, p. 195-210) securitization is closely related to the emergence of the financial crisis, Gebhard (2016, p. 169) state that also credit risk played a big role in the start of the financial crisis, and Allen and Powell (2011) argue that the financial crisis highlighted the increasing importance of credit risk measurement. Therefore the first hypothesis will be:
H1: Banks that are involved in securitization will have higher credit risk compared to banks that are not involved in securitization in the period before the financial crisis.
This hypothesis can confirm one of the conclusions drawn by Casu et al. (2011). They state that securitizing banks hold riskier assets compared to non-securitizing banks.
The financial crisis caused a big shock in finance and banks were forced to change the way they worked (Atkinson et al., 2013). Because of the known relation between securitization, credit risk and the financial crisis, the second hypothesis is:
H2: The effect of securitization on the credit risk of banks will be lower during the financial crisis compared to before the financial crisis.
Based on their research Berglund and Mäkinen (2016) conclude that when a crisis occurs, banks learn from their mistakes and they adjust their business models. Because of the known cause of the financial crisis, the third hypothesis of this research is:
H3: The effect of securitization on the credit risk of banks will be lower after the financial crisis compared to before the financial crisis.
14
4.
Methodology
In this chapter the methodology will be discussed which is used to perform the empirical tests to investigate the effect of securitization on credit risk in the three different timeframes. The first section discusses the sample which is used. Thereafter the dependent variables RWATA and charge-off ratio are described, which is followed by the explanation of the independent variables in the third section.
4.1 Sample
The aim of this thesis is to investigate the effect of securitization on a bank’s credit risk over three different timeframes. There are several studies in which credit risk had to be measured. By measuring credit risk, I will mainly follow Casu et al. (2011), Aggarwal and Jacques (2001), Avery and Berger (1990), Berger (1995), Berger and Udell (1994), and Shrieves and Dahl (1992). The data that is used is Bank Holding Company data from Y-9C reports for US banks. All Bank Holding Companies filled in these reports each quarter of a year and these are gathered by the Federal Reserve Bank of Chicago. The data from Bank Holding Companies is used because within a group the administration of capital and risk activities is normally done by the holding company. Demsetz and Stahan (1997), Jiangli and Pritsker (2008), Jiangli et al. (2007) and Casu et al. (2011) also used these reports by investigating securitization.
In the Y-9C reports consolidated financial data is collected that includes balance sheet, income statement, supporting schedules and schedules about off-balance-sheet items. The data that will be used in this research will start in 2003 up to and including 2017. The first quarter of 2003 to the fourth quarter of 2017 results in 64 quarters in total. As mentioned before, I will use three different timeframes that represent the period before the crisis, during the crisis and after the crisis. The period before the financial crisis is represented by the first quarter of 2003 up to and including the second quarter of 2008 when the crisis started (Foster, 2009, p.11). I will use the first quarter of 2013 as the first period after the financial crisis, since five to six years after the start of the financial crisis, the US reached the highest point in capita income in 2007-2008 (Reinhart and Rogoff, 2014, p. 50-55).
The data is quarterly available, so each dataset has to be merged per period to get the right database for each period. When an observation misses information about total assets, securitization activities, leverage or non-performing loans at any time in the sample period,
15 the observation is removed from the dataset. In table 1 an overview of the data used per period. These variables will be discussed intensively later on.
In the first period before the financial crisis the final sample has 36,136 observations. The number of observations in this period that were involved in securitization is 1,926, which is 5.33%. In the second period during the financial crisis the total observation is 18,349, which exist of 1,346 banks that were involved in securitization and this is 7.34%. In the period after the financial crisis there are 15,983 observations in the final sample. There are 1,104 securitizing banks which is 6.9%.
4.2 Dependent variables
To investigate the effects of securitization on bank risk over the different timeframes, I will use credit risk as dependent variable. Credit risk is measured by risk-weighted assets to total assets and by charge-off ratio. In the decisions about the dependent and independent variable are mainly Casu et al (2011) and Cebenoyan and Strahan (2004) followed. Casu et al. (2011) did a research that is about the same specific effects of securitization and Cebenoyan and Strahan investigated more the risk component of this research.
4.2.1 RWATA
There are a lot of prior studies that used credit risk in their research. This study will follow the same variable that is used by Avery and Berger (1990), Shrieves and Dahl (1992), Berger and Udell (1994), Berger (1995), Aggarwal and Jacques (2001) and Casu et al. (2011) to measure Table 1. Sample selection
Sample selection procedure
Number of observations 2003Q1 – 2008Q2 2008Q3 – 2012Q4 2013Q1 – 2017Q4
Initial Sample 84,599 60,175 60,142
Remove missing data total assets 47,797 41,316 43,045
Remove missing data securitization 119 0 12
Remove missing data leverage 547 505 1,094
Remove missing data NPL 0 5 8
16 credit risk. This is the ratio between risk-weighted assets and total assets. This is called RWATA in the empirical tests. RWATA is calculated as follows: With the Basel I regulations, a bank’s assets and off-balance-sheet activities are divided into four groups based on the level of the credit risk of the asset. Each group get a weight based on the relative risk of the assets. First group are the assets with zero default risk. These are for example government securities and they have a risk-weight of zero. The second group are the low-risk assets, for example interbank deposits and have a risk weight of 0.2. Third group includes assets with average default risk, like mortgages and have a risk-weight of 0.5. The fourth group exist of high-risk assets and have a risk weight of one. These are commercial loans. The formula of RWA is therefore: RWA = 0 * Group 1 + 0.2 * Group 2 + 0.5 * Group 3 + 1 * Group 4. According to Avery and Berger (1990) there is correlation between risky behaviour at banks and relative risk weight. This correlation includes information that predicts the problems in future performance of banks, like credit defaults. Both variables risk-weighted assets and total assets are represented in the Y-9C reports.
4.2.2. Charge-off Ratio
The second dependent variable I use for credit risk is charge-off ratio. Because the asset risk is not only caused by the risk categories of a bank, I use charge-off ratio as a proxy for asset quality. This measures the quality of the assets and is also used by Jiangli and Pritsker (2008) and Casu et al. (2011). This ratio is shows the relative charge-offs of a bank in a specific period. I will use the net charge-off ratio, which indicates that the recoveries on loans and leases are deducted from the charge-offs on loans and leases. To get a ratio this amount has to be divided by the average total loans and leases for that period. This leaves us to the following calculation:
(Total Charge-Offs on Loans and Leases - Total Recoveries on Loans and Leases) / Quarterly average loans and leases
Casu et al. (2011) used the same variables and calculation to get to this variable in their research. The variables that are needed in the calculation are available in the Y-9C reports. Because RWATA measures the allocation among risk categories and the charge-off ratio measures the asset quality, these two variables as dependent variables complement each other and are a good choice to use as proxies for credit risk.
17 4.3 Independent variables
The independent variables that will be used are explained in this section of the methodology. The empirical tests will include one dummy and six control variables. The control variables in the tests are bank size, leverage, liquidity ratio, return on assets, capital ratio and non-performing loans.
First, the securitization dummy (SEC) equals 1 if the bank is involved in securitization in one of the observations in the specific timeframe and 0 if the bank is not involved in securitization in any of the observations in the timeframe. The securitization activities are available in the Y-9C reports and are divided in the categories; Securitization activity 1–4 Family Residential Loans, Securitization activity Home Equity Lines, Securitization activity Credit Card Receivables, Securitization activity Auto Loans, Securitization activity Other Consumer Loans, Securitization activity Commercial and Industrial Loans and Securitization activity All Other Loans and Leases.
The first control variable that is used is the logarithm of Bank size (SIZE). According to Jiangli et al. (2008), Minton et al. (2008) and Uzun and Webb (2007) bank size is important in the decision whether or not to engage in securitization. They argue that securitization comes with high fixed costs, which is easier to bear for large banks because they can enjoy economics of scale and can diversify their risk (Hughes, Mester and Moon (2001). Therefore one could expect that there is a positive relation between bank size and involving in securitization. The variable total assets is available in the Y-9C reports. This variable is also used by Cebenoyan and Strahan (2004), Jiangli et al. (2007), Le et al. (2016), Loutskina and Strahan (2009), Casu et al. (2011), Stiroh (2006), Jiangli and Pritsker (2008) and Baele et al. (2011).
I will follow Jiangli et al. (2007), Loutskina and Strahan (2009), Stiroh (2006a) and Jiangli and Pritsker (2008) by using leverage as control variable. According to Jiangli and Pritsker (2008), securitization of mortgages is used to increase leverage. Tier 1 leverage ratio is calculated by the following formula:
Tier 1 leverage ratio = Tier 1 capital / (average total consolidated assets + certain off-balance sheet exposures)
18 In the empirical tests, leverage is measured by 1/ Tier 1 leverage ratio. Tier 1 leverage capital ratio is represented in the Y-9C reports with code BHCK7204 and BHCA7204. Because from 2015 the Tier 1 leverage capital ratio is measured with another variable, BHCKA7204, these two variables are merged in the third period. The variable is winsorized at a 1% level to prevent for outliers. Since I divide 1 by Tier 1 leverage ratio, an increase in this variable indicates that the leverage of the bank decreases.
The third control variable in the regression is the liquidity ratio (LIQ). By choosing liquidity ratio as control variable Cebenoyana and Strahan (2004), Loutskina and Strahan (2009), Baele et al. (2011) and Casu et al. (2011) are followed. The variable is calculated by liquid assets divided by total assets. Liquidity ratio is calculated with the following formula:
Liquidity ratio = (Cash and due from depository institutions + Trading assets + Held-to-maturity securities + Available-for-sale securities + Federal funds sold and securities purchased under agreements to resell - Federal funds purchased and securities sold under agreements to repurchase) / Total Assets
All the variables in de calculation are available in the Y-9C reports. The variable is winsorized at a 1% level to prevent for outliers.
Further, return on assets (ROA) is the next control variable in the empirical tests. This control variable is used by following Cebenoyan and Strahan (2004) and Casu et al. (2011). According to Cebenoyan and Strahan (2004) and Purnanandam (2009) it can be expected that securitizing banks will have a higher return on assets compared to non-securitizing banks, because securitizing banks have additional liquidity to invest for higher returns. The variable is calculated by the following formula:
Return on Assets = Net income (loss) / Total consolidated assets * 100
Both Net income (loss) and Total consolidated assets are available in the Y-9C reports. The variable Return on Assets is winsorized at a 1% level to prevent for outliers.
Capital ratio (CAPR) is used as the fifth control variable. Cebenoyan and Strahan (2004), Stiroh (2006), Baele et al. (2011) and Casu et al. (2011) also used this ratio as variable.
19 Capital ratio is represented in the Y-9C reports as Tier 1 risk-based capital ratio. In the empirical tests the capital ratio is divided by 100. It is winsorized at a 1% level to prevent for outliers.
The last control variable that is used in the regression is non-performing loans ratio (NPL). This ratio is also used by Casu et al. (2011), Cebenoyan and Strahan (2014), Baele et al. (2011) and Stiroh (2006) and is called NPL in the empirical tests. The following formula represents the calculation of this variable:
Non-performing loans ratio = Ln (Total past due 90 days or more and still accruing + Total nonaccrual loans, leases, and other assets - Debt Securities and Other Assets past due 90 days or more and still accruing - Debt Securities and Other Assets Nonaccrual) / Quarterly average loans and leases.
20
5.
Empirical Design
In this part the methodology will be discussed. The research question that is central in this paper is the effect of securitization on a bank’s credit risk in three different periods. There are three hypotheses that will be tested.
The following equations are used to do a regression of securitization on the proxies for credit risk:
RWATA = + 1SECi + 2SIZEi + 3LEVi + 4LIQi + 5ROAi + 6CAPRi + 7NPLi
Charge-off ratio = + 1SECi + 2SIZEi + 3LEVi + 4LIQi + 5ROAi + 6CAPRi + 7NPLi
Both regressions will be used in the three hypotheses which are discussed in chapter 3. In these regressions RWATA and Charge-off ratio are the dependent variables which measure the credit risk of a bank. SEC is the dummy that is used to show if the bank is a non-securitizing or a non-securitizing bank. LEV, SIZE, LIQ, ROA, CAPR and NPL are control variables. To prevent for outliers the variables leverage, liquidity ratio, return on assets, capital ratio, non-performing loans, ratio between risk-weighted assets and total assets and charge-off ratio are winsorized at a 1% level.
In both regressions and all hypotheses, the coefficient of interest is 1. The expectation for the
first hypothesis is that both 1 will be positive. This because previous research suggests that a
securitizing bank will hold riskier assets (Casu et al., 2011) and that securitization will have a positive effect on a bank’s risk (Dionne and Harchaoui, 2003). For the second and third hypothesis I expect 1 be lower than in the first period, or even negative. This based on prior
research of Atkinson et al. (2013) who conclude that after the financial crisis banks changed the way they worked and Allen and Powell (2011) argue that the importance of credit risk increased after the financial crisis.
21
6.
Data and Descriptive Statistics
The sixth chapter, classified as the data and descriptive statistics, provides an overview of the used variables and its values summarized. For every period the sample is divided in observations for securitizing and non-securitizing banks. Further, the correlation for all observations together of the dependent, independent and control variables will be elaborated.
6.1 Descriptive Statistics
Tables 2 to 7 show the summary statistics for all the used variables. In tables 2 and 3, the statistics for the Bank Holding Companies for the period from the first quarter of 2003 up to the second quarter of 2008 are shown. The summary statistics for the period from the third quarter of 2008 up to the last quarter of 2012 are illustrated in tables 4 and 5. In tables 6 and 7 are the statistics for the period from the first period of 2013 until the last period of 2017 shown. Since securitization is an activity that recurs, I divided the Bank Holding Companies in two groups. When a Bank Holding Company is involved in securitization in one of the periods of the timeframe, the bank is assigned to the group of securitizing banks in that particular timeframe. Banks with no securitization activities in a period of a timeframe are assigned to the non-securitizing banks.
6.1.1. Descriptive Statistics before financial crisis
Table 3 shows 1926 observations of securitizing banks before the financial crisis and table 2 shows 34210 observations of non-securitizing banks. This indicates that 5.33% of all the observations in this period is a securitizing bank. The mean differences between tables 2 and 3 are represented in table 3 and are all significant at 1% level.
The first big difference that stands out between table 2 and table 3 is the difference in bank size between securitizing and non-securitizing banks. Jiangli et al. (2008), Minton et al. (2008) and Uzun and Webb (2007) already stated that this is to be expected because of the high fixed costs that occur when a bank involves in securitization. Also the higher return on assets for securitizing banks is in line with prior studies of Jiangli et al. (2007) and Casu et al (2011). They state that securitizing banks improve their profitability because they have riskier and more profitable loans in their portfolios. The tables also tell us that securitizing banks have a lower liquidity ratio. This is explainable because securitizing banks have a better access to external funding and therefore they need less liquidity than non-securitizing banks. Capital ratio of securitizing banks is lower than the capital ratio of non-securitizing banks.
22 This is in line with findings of Cebenoyan and Strahan (2004) and Minton et al. (2008), who concluded that banks that engage in management of credit risk will hold less capital. Risk-weighted assets relative to total assets, charge-off ratio and non-performing loans are variables used in this research which can say something about the risk of banks. Looking at table 3, these variables are all higher for securitizing banks compared to non-securitizing banks. This finding in statistics is in line with the results of Jiangli and Pritsker (2008), Minton et al. (2008) and Casu et al. (2011).
Table 2. Descriptive statistics before financial crisis
Descriptive Statistics; Non-securitizing Bank Holding Companies 2003Q1-2008Q2
Variables Obs. Mean Median Std. Dev. Min P25 P75 Max
Bank size 34210 6.318 6.110 1.085 3.270 5.508 6.796 13.241 Leverage 34210 11.364 11.287 2.563 5.071 9.775 12.804 19.342 Liquidity ratio 34210 .252 .230 .130 .033 .156 .329 .642 ROA 34210 .650 .565 .464 -.389 .307 .912 2.273 Capital ratio 34210 .130 .117 .048 .064 .101 .144 .353 NPL 34210 .009 .006 .010 0 .003 .011 .056 RWATA 34210 .727 .737 .117 .390 .654 .809 .980 Charge-off ratio 34210 .001 .001 .002 -.001 .0001 .001 .014 Notes: This table shows summary statistics for the main variables used in this research for the period from the first quarter of 2003 up to and including the second quarter of 2008. These statistics are for non-securitizing Bank Holding Companies. All variables are defined in Appendix A. Bank size, leverage, liquidity ratio, return on assets, capital ratio and non-performing loans (NPL) are control variables. Risk-weighted assets to total assets (RWATA) and Charge-off ratio are dependent variables. *** p<0.01, ** p<0.05, *p<0.1.
23 6.1.2. Descriptive Statistics during financial crisis
Table 4 and 5 below are the summary statistics of non-securitizing and securitizing banks respectively during the financial crisis, which is the period starting from the third quarter of 2008 up to and including the fourth quarter of 2012. Table 5 shows that in this period 1,346 banks were involved in securitization of the total 18,349 banks, which is 7.33%.
Compared to the summary statistics of the period before the financial crisis, bank size has still a high mean difference which is already stated by Jiangli et al. (2008), Minton et al. (2008) and Uzun and Webb (2007). Table 5 also shows an increased mean difference of charge-off ratio compared to the period before the crisis, which suggests a lower asset quality for securitizing banks. It is also remarkable that non-securitizing banks have a higher mean of RWATA. This suggests in line with Atkinson et al. (2013) that securitizing banks invest in less risky assets during the financial crisis. Other links to literature are given in the previous paragraph.
Table 3. Descriptive statistics before financial crisis
Descriptive Statistics; Securitizing Bank Holding Companies 2003Q1-2008Q2
Variables Obs. Mean Median Std. Dev. Min P25 P75 Max Mean diff. Mean diff. %
Bank size 1926 9.057 9.181 2.464 4.984 6.657 10.94 14.673 2.739*** 43.352 Leverage 1926 12.457 12.516 3.221 5.071 10.858 14.368 19.342 1.093*** 9.618 Liquidity ratio 1926 .226 .180 .138 .033 .126 .297 .642 -.026*** -10.317 ROA 1926 .710 .605 .550 -.389 .315 1.001 2.273 .060*** 9.231 Capital ratio 1926 .116 .102 .052 .064 .085 .126 .353 -.014*** -10.769 NPL 1926 .011 .008 .010 0 .004 .014 ..056 .002*** 22.222 RWATA 1926 .744 .767 .135 .390 .659 .843 .980 .017*** 2.338 Charge-off ratio 1926 .002 .001 .003 -.001 .0004 .003 .014 .001*** 100 Notes: This table shows summary statistics for the main variables used in this research for the period from the first quarter of 2003 up to and including the second quarter of 2008. These statistics are for securitizing Bank Holding Companies. All variables are defined in Appendix A. Bank size, leverage, liquidity ratio, return on assets, capital ratio and non-performing loans (NPL) are control variables. Risk-weighted assets to total assets (RWATA) and Charge-off ratio are dependent variables.
24 Table 4. Descriptive statistics during financial crisis
Descriptive Statistics; Non-securitizing Bank Holding Companies 2008Q3-2012Q4
Variables Obs. Mean Median Std. Dev. Min P25 P75 Max
Bank size 17003 7.087 6.802 1.032 4.364 6.433 7.436 13.649 Leverage 17003 11.691 10.989 8.827 -40.161 9.551 12.739 59.172 Liquidity ratio 17003 .251 .236 .117 .043 .167 .317 .623 ROA 17003 .128 .272 1.033 -5.018 .028 .608 2.158 Capital ratio 17003 .129 .124 .053 -.018 .103 .147 .379 NPL 17003 .036 .024 .036 .0007 .012 .045 .200 RWATA 17003 .715 .724 .111 .376 .646 .790 .959 Charge-off ratio 17003 .007 .003 .010 -.0004 .001 .008 .055 Notes: This table shows summary statistics for the main variables used in this research for the period from the third quarter of 2008 up to and including the fourth quarter of 2012. These statistics are for non-securitizing Bank Holding Companies. All variables are defined in Appendix A. Bank size, leverage, liquidity ratio, return on assets, capital ratio and non-performing loans (NPL) are control variables. Risk-weighted assets to total assets (RWATA) and Charge-off ratio are dependent variables. *** p<0.01, ** p<0.05, *p<0.1.
Table 5. Descriptive statistics during financial crisis
Descriptive Statistics; Securitizing Bank Holding Companies 2008Q3-2012Q4
Variables Obs. Mean Median Std. Dev. Min P25 P75 Max Mean diff. Mean diff. %
Bank size 1346 9.273 8.925 2.435 5.558 7.053 11.143 14.679 2.186*** 30.845 Leverage 1346 11.919 11.13 11.327 -40.161 9.443 13.333 59.172 .228 1.950 Liquidity ratio 1346 .243 .218 .126 .043 .153 .297 .623 -.008** -3.187 ROA 1346 .148 .288 1.076 -5.018 .009 .627 2.158 .020 15.625 Capital ratio 1346 .127 .120 .060 -.017 .102 .146 .379 -.002 -1.550 NPL 1346 .042 .031 .038 .0007 .017 .055 .201 .006*** 16.667 RWATA 1346 .706 .722 .129 .376 .631 .788 .958 -.009*** -1.259 Charge-off ratio 1346 .010 .006 .012 -.0005 .002 .013 .055 .003*** 42.857 Notes: This table shows summary statistics for the main variables used in this research for the period from the first quarter of 2003 up to and including the second quarter of 2008. These statistics are for securitizing Bank Holding Companies. All variables are defined in Appendix A. Bank size, leverage, liquidity ratio, return on assets, capital ratio and non-performing loans (NPL) are control variables. Risk-weighted assets to total assets (RWATA) and Charge-off ratio are dependent variables. *** p<0.01, ** p<0.05, *p<0.1.
25 6.1.3. Descriptive Statistics after financial crisis
The descriptive statistics for the period after the financial crisis are presented in tables 6 and 7 below. According to Reinhart et al. (2014) the economy was recovered in that timeframe and this period contains the first quarter of 2013 up to and including the last quarter of 2017. In the table are 1,104 observations of securitizing banks of the total 15,983 observations. This means that 6.9% of all the observations is a securitizing bank. This is less than the second period, however this is not not a big change.
It is striking that the mean of charge-off ratio of securitizing banks is more than 85% higher than non-securitizing banks, which indicates that securitizing banks have a lower asset quality. The other differences between the two groups are not outstanding. Links to literature are given in the paragraph of the period before the crisis.
Table 6. Descriptive statistics after financial crisis
Descriptive Statistics; Non-securitizing Bank Holding Companies 2013Q1-2017Q4
Variables Obs. Mean Median Std. Dev. Min P25 P75 Max
Bank size 14879 7.496 7.219 1.138 2.857 6.733 7.983 13.047 Leverage 14879 5.449 6.398 5.640 .0004 .001 10.246 22.321 Liquidity ratio 14879 .261 .239 .127 .056 .167 .329 .681 ROA 14879 .563 .483 .460 -.591 .250 .784 2.504 Capital ratio 14879 .144 .133 .052 .033 .116 .159 .394 NPL 14879 .014 .009 .017 .0002 .005 .017 .112 RWATA 14879 .719 .728 .120 .353 .648 .802 .971 Charge-off ratio 14879 .139 .051 .285 -.192 .006 .156 1.895
Notes: This table shows summary statistics for the main variables used in this research for the period from the first quarter of 2013 up to and including the fourth quarter of 2017. These statistics are for non-securitizing Bank Holding Companies. All variables are defined in Appendix A. Bank size, leverage, liquidity ratio, return on assets, capital ratio and non-performing loans (NPL) are control variables. Risk-weighted assets to total assets (RWATA) and Charge-off ratio are dependent variables. *** p<0.01, ** p<0.05, *p<0.1.
26 6.2 Pearson Correlation Matrix
Table 7 gives an overview of the Pearson correlation between the dependent, independent and control variables that are used in this thesis. This analysis is performed because of the chance of multicollinearity between two or more variables. Multicollinearity means that it is possible to linearly predict two independent variables by each other, which results in inferences that are unreliable. Tay (2017) states that when a coefficient between two variables is 0,8 or more, it states multicollinearity. It can also be used as optimization criteria for deriving noise reduction (Benesty et al., 2009). The correlation matrix shows no correlation above 0.8, however there are some variables that are relatively high correlated. The variables Bank size and Securitization are relatively high correlated with significant coefficient of 0.433. This correlation is already predicted by Jiangli et al. (2008), Minton et al. (2008) and Uzun and Webb (2007) because of the high fixed costs. Capital ratio and Leverage are also relatively highly correlated with a significant coefficient of -0.445. This is explainable, since capital ratio and leverage ratio are calculated with capital divided by risk-weighted assets and total assets respectively. Two other variables that are relatively highly correlated are risk-weighted assets to total assets and liquidity ratio. This is also explainable, since both depend on total assets. Furthermore, liquidity ratio and capital ratio, return on assets and non-performing loans, non-performing loans and charge-off ratio, and capital ratio and RWATA are relatively high correlated.
Table 7. Descriptive statistics after financial crisis
Descriptive Statistics; Securitizing Bank Holding Companies 2013Q1-2017Q4
Variables Obs. Mean Median Std. Dev. Min P25 P75 Max Mean diff. Mean diff.
Bank size 1104 10.062 9.741 2.187 5.902 8.168 11.803 14.762 2.566*** 34.232 Leverage 1104 3.508 .001 5.08 .0004 .001 8.745 22.321 -1.941*** -35.621 Liquidity ratio 1104 .248 .222 .127 .056 .164 .295 .681 -.013*** -4.981 ROA 1104 .596 .518 .496 -.591 .263 .822 2.504 .033** 5.861 Capital ratio 1104 .144 .128 .051 .033 .117 .158 .394 0 0 NPL 1104 .019 .012 .022 .0002 .007 .020 .112 .005*** 35.714 RWATA 1104 .725 .739 .117 .353 .648 .813 .971 .006 .834 Charge-off ratio 1104 .258 .116 .406 -.192 .028 .283 1.895 .119*** 85.612 Notes: This table shows summary statistics for the main variables used in this research for the period from the first quarter of 2003 up to and including the second quarter of 2008. These statistics are for securitizing Bank Holding Companies. All variables are defined in Appendix A. Bank size, leverage, liquidity ratio, return on assets, capital ratio and non-performing loans (NPL) are control variables. Risk-weighted assets to total assets (RWATA) and Charge-off ratio are dependent variables. *** p<0.01, ** p<0.05, *p<0.1.
27 Table 8. Pearson Correlation Matrix
Pearson Correlation Matrix
Securitization Bank Size Leverage Liq. Ratio ROA Capital Ratio NPL RWATA Ch.-off ratio Securitization 1 Bank Size 0.433*** 1 Leverage -0.005 -0.158*** 1 Liq. Ratio -0.033*** 0.0587*** -0.034*** 1 ROA 0.006 -0.0359*** 0.218*** 0.061*** 1 Capital Ratio -0.033*** -0.0832*** -0.445*** 0.482*** 0.242*** 1 NPL 0.06*** 0.0794*** 0.187*** -0.014*** -0.463*** -0.057*** 1 RWATA 0.011** 0.0315*** -0.035*** -0.749*** -0.023*** -0.525*** -0.041*** 1 Ch.-off ratio 0.086*** 0.125*** 0.171*** -0.024*** -0.430*** -0.072** 0.556*** 0.0237*** 1 Notes: This table provides an overview of the Pearson Correlation Matrix of all the dependent and independent variables for all observations from 2003Q1-2017Q4. This sample exists of all securitizing and non-securitizing banks and has in total 70,468 observations. The sample is generated from the FR Y-9C reports for the period 2003Q1 through 2008Q2. All variables are defined in Appendix A. *** p<0.01, ** p<0.05, * p<0.1.
28
7.
Empirical Findings
In the previous chapters the methodology and data is discussed to perform the empirical tests. In this chapter the results of these tests will be discussed extensively. The results of each timeframe will be discussed in a section. The results of the regressions on RWATA and charge-off ratio will be interpreted and will be related to credit risk followed by a brief overview of some remarkable control variables.
7.1 Before the financial crisis
Table 9 below illustrates the results of the regressions of the period before the financial crisis. Both are OLS regressions with RWATA and Charge-off ratio as dependent variables. Both regressions exist of a sample with 36,136 observations, of which 1,926 are securitizing banks. The numbers of interests are the coefficients of the securitization dummy, which are tested with RWATA and Charge-off ratio as dependent variable.
The securitization dummy has a positive value of 0.0099 with a standard error of 0.0016 in the regression that contains RWATA as independent variable and is significant at a 1% level. Because the variable Securitization represents 0 when a bank is a non-securitizing bank and 1 if the bank is involved in securitization. Therefore the results indicate that a bank has a higher RWATA of 0.0099 when the bank is involved in securitization compared to a non-securitizing bank and this is in line with the findings of Casu et al. (2011). An increasing value of RWATA means that a bank has relatively more risk-weighted assets. The means of RWATA in tables 2 and 3 are between 0.72 and 0.75. Therefore the effect of 0.01 is relatively low. This result indicates a positive relation between securitization and risk-weighted assets, however the impact of securitization on risk-weighted assets to total assets is relatively low. This supports hypothesis 1 that states that banks that are involved in securitization will have higher credit risk than banks that are not involved in securitization in the period before the financial crisis.
Within the regression that includes charge-off ratio as independent variable, the securitization dummy has a positive coefficient of 0.0006, which is significant at the 1% level. This means that when a bank is involved in securitization, the charge-off ratio is 0.0006 higher compared to not involving in securitization. The means of charge-off ratio in tables 2 and 3 are 0.001 and 0.002. Therefore an effect of 0.0006 is relatively high. Charge-off ratio is calculated by the ratio between the total charge-offs minus recoveries divided by the total loans and leases. This positive relation between securitization and charge-off ratio therefore
29 indicates that a bank that is involved in securitization will have more charge-offs on loans and leases relative to the total loans and leases compared to not involving in securitization. This finding is in line with expectations and therefore supports hypothesis 1 that states that banks that are involved in securitization will have higher credit risk than banks that are not involved in securitization in the period before the financial crisis.
Looking at the control variables in both regressions all the coefficients are significant at 1% level except liquidity ratio in the regression on charge-off ratio. Leverage is calculated with 1 divided by Tier 1 leverage ratio and it can be seen that leverage has a negative coefficient of 0.0228 in the regression on RWATA. This indicates that when the leverage of a bank decreases, the risk-weighted assets to total assets also decrease. The effect between leverage and charge-off ratio is in this regression vice versa and it is striking that this effect is that small. This is contrary to Jiangli and Pritsker (2008), who state that securitization is used to increase leverage. The negative coefficient of bank size indicates that when the logarithm of bank size increases with 1 a bank RWATA decreases with 0.002. Furthermore, an increase of return on assets and non-performing loan will result in higher RWATA and liquidity ratio and capital ratio vice versa. An increase in return on assets, capital ratio and non-performing loans will result in a higher charge-off ratio and an increase of liquidity ratio will result in a lower charge-off ratio.
RWATA and charge-off ratio are both proxies for credit risk and have both a positive relation with securitization and this indicates that a securitizing bank has a higher credit risk than when it not involves in securitization. Therefore I conclude from these regressions that when banks are involved in securitization, they will have higher credit risk compared to not involve in securitization before the start of the financial crisis. This supports hypothesis 1 and is in line with previous research about the effect of securitization on bank risk done by Jiangli and Pritsker (2008), Minton et al. (2008) and Casu et al. (2011).
30 7.2 During the financial crisis
Table 10 presents the regression output for the period during the financial crisis. Hypothesis 2 is tested with the same models as Hypothesis 1. The sample for this period consists of 18,349 observations, of which 1,346 are securitizing banks. The numbers of interests are the coefficients of the securitization dummy, which are tested with RWATA and Charge-off ratio as dependent variable.
The coefficient of the securitization dummy has a negative value of 0.0052 with a standard error of 0.0024 in the regression with RWATA and is significant at a 5% level. This indicates a bank that is involved in securitization will have a lower RWATA of 0.0052 compared to if the bank was not involved in securitization. This decreasing value of RWATA
Table 9. Before the Financial Crisis Regression on RWATA and Charge-off ratio
(1) (2)
Variables RWATA Charge-off ratio
Securitization 0.0099*** 0.0006*** (0.0016) (5.33e-05) Leverage -0.0228*** 6.75e-05*** (0.0002) (6.60e-06) Bank size -0.0020*** 0.0001*** (0.0002) (9.06e-06) Liquidity ratio -0.4210*** -0.0001 (0.0033) (0.0001) Return on Assets 0.0149*** 0.0005*** (0.0007) (2.34e-05) Capital ratio -1.6840*** 0.0024*** (0.0126) (0.0004) NPL 0.3900*** 0.0900*** (0.0327) (0.0011) Constant 1.3110*** -0.0019*** (0.0037) (0.0001) Observations 36,136 36,136 R-squared 0.7480 0.1760
An OLS Regression is performed with risk-weighted assets to total assets (RWATA) and Charge-off ratio as dependent variables. The sample is generated from the FR Y-9C reports for the period 2003Q1 through 2008Q2. All variables are defined in Appendix A. Standard error in parentheses. *** p<0.01, ** p<0.05, * p<0.1.
31 means that a bank has relatively less risk-weighted assets to total assets when it involves in securitization. The means of RWATA of the observations during the financial crisis are 0.715 for non-securitizing banks and 0.706 for securitizing banks. Therefore a negative coefficient of 0.0052 is relatively low. However, the result is in line with hypothesis 2, which states that the effect of securitization on the credit risk of banks compared to non-securitizing banks will be lower during the financial crisis than before the financial crisis, since the coefficient is significant and even negative.
Looking at the charge-off ratio the coefficient of the securitization dummy shows a positive value of 0.0007, a standard error of 0.0002 and is significant at 1% level. Therefore a securitizing bank has a higher charge-off ratio of 0.0007 compared to not involve in securitization. The means of charge-off ratio of the observations during the financial crisis for securitizing banks and non-securitizing banks are 0.007 and 0.01 respectively, which makes an increase of 0.0007 look relatively mediocre. Similar to the previous section, this relation between securitization and charge-off ratio indicates that a bank that is involved in securitization will have more charge-offs on loans and leases relative to the total loans and leases compared to not involving in securitization. Hypothesis 2 states that the effect of securitization on credit risk of banks compared to non-securitizing banks will be lower during the financial crisis than before the financial crisis. Examining this result, it can be concluded that based on this finding there is no support for hypothesis 2.
Table 10 below shows that all the control variables in the regression of RWATA are significant at a 1% level and in the regression of charge-off ratio only leverage and liquidity ratio are not significant. Compared to the control variables in the test for the period before the financial crisis liquidity ratio, return on assets and non-performing loans differ. The coefficient of return on assets are both negative, which indicates that a bank with less return on assets has relatively less risky assets and relatively less charge-offs. This is the only control variable in these regressions with no opposite results.
It is remarkable that during the financial crisis securitizing banks have relatively less risky assets and a higher charge-off ratio. This indicates that banks that are involved in securitization invest in less risky assets, but have lower quality of assets. Hence, it is hard to say if securitizing banks had a higher credit risk during the financial crisis compared to non-securitizing banks. Therefore it can be concluded that these results support hypothesis 2 that state that the effect of securitization on the credit risk of banks will be during the financial crisis compared to before the financial crisis. These results are in line with Atkinson et al.
32 (2013) who states that banks were force to change the way they worked after the emergence of the financial crisis.
7.3 After the financial crisis
The results of the regression on RWATA and charge-off ratio for the period after the financial crisis are illustrated in table 10 below. The sample in this period consists of 15,983 observations, of which 1104 observations of securitizing banks. These regressions to examine hypothesis 3 are the same as used for hypothesis 1 and 2, however this sample is for the period after the financial crisis.
The dummy for securitization has a positive value of 0.0005 with a standard error of 0.0027 in the regression that contains RWATA as dependent variable. This coefficient is not significant and therefore there is no significant effect found between securitization and
risk-Table 10. During the Financial Crisis Regression on RWATA and Charge-off ratio
(1) (2)
Variables RWATA Charge-off ratio
Securitization -0.0052** 0.0007*** (0.0024) (0.0002) Leverage -0.0008*** 2.57e-06 (6.35e-05) (6.35e-06) Bank size -0.0035*** 0.0008*** (0.0005) (4.67e-05) Liquidity ratio -0.5900*** 0.0003 (0.0051) (0.0005) Return on Assets -0.0052*** -0.0051*** (0.0007) (6.65e-05) Capital ratio -0.4690*** 0.0120*** (0.0124) (0.0012) NPL -0.2620*** 0.0612*** (0.0178) (0.0018) Constant 0.9670*** -0.0020*** (0.00388) (0.0004) Observations 18,349 18,349 R-squared 0.550 0.427
Regression of risk-weighted assets to total assets (RWATA) and Charge-off ratio. The sample is generated from the FR Y-9C reports for the period 2008Q3 through 20012Q4. All variables are defined in Appendix A. Standard error in parentheses. *** p<0.01, ** p<0.05, * p<0.1.
33 weighted assets to total assets. Therefore based on this finding there is no support or rejection for hypothesis 3.
Looking at the securitization dummy in the regression that contains charge-off ratio as dependent variable the positive coefficient of 0.0491 indicates that a bank that involved in securitization has a higher charge-off ratio compared to not involve in securitization. The coefficient is significant at 1% level. This means that they have more charge-offs on loans and leases relative to total loans and leases compared to not involving in securitization. The means of charge-off ratio are for securitizing and non-securitizing banks 0.139 and 0.258 respectively, which makes an increase of 0.0491 relatively high. Since this coefficient is significantly higher than before the financial crisis, this finding supports hypothesis 3 that states that the effect of securitization on the credit risk of banks compared to non-securitizing will be lower after the financial crisis than before the financial crisis.
Looking at the control variables in both regressions all the coefficients are significant at 1% level. It is striking that the coefficient of leverage in the charge-off regression is many times higher compared to its value before the financial crisis. Furthermore, all the control variables are comparable with their value before the financial crisis, only the coefficient of non-performing loans in the RWATA regression is negative instead of positive. Just like in the other two periods the coefficient of bank size is negative which indicates that RWATA decreases 0.0025 with an increase of 1 in terms of logarithm bank size.
The regression on charge-off ratio supports hypothesis 3 in that securitizing banks have higher asset quality. On the other hand, in the regression on RWATA there is no relation found between securitization and credit risk. Therefore I found no support for hypothesis 3 and I do not conclude that the effect of securitization on the credit risk of banks is lower after the financial crisis compared to before the financial crisis, although the positive effect on charge-off ratio suggests a positive effect.
34 7.2 Robustness Test
The point estimates do not change by adding robust standard errors, however the standard errors do change a little for the regressions for the period before the financial crisis. Table 11 shows the regressions of this research with robust standard errors. Compared to table 9, which are the regressions of the period before the crisis, the robust standard errors are higher than the standard error for the regressions of RWATA and for the charge-off ratio. The coefficients of the securitization dummy are still significant at a 1% level with robust standard errors.
Looking at the robust standard errors for the period during the financial crisis, these are also higher than the standard errors in table 9. The robust standard error of the RWATA regression is slightly higher and the coefficient is significant at a 10% level instead of the 5% level of the normal standard error. The coefficient of the charge-off ratio regression is still significant at the 1% level.
Table 11. After the Financial Crisis Regression on RWATA and Charge-off ratio
(1) (2)
Variables RWATA Charge-off ratio
Securitization 0.0005 0.0491*** (0.0027) (0.0097) Leverage -0.0031*** 0.0062*** (0.0001) (0.0004) Bank size -0.0025*** 0.0196*** (0.0005) (0.0018) Liquidity ratio -0.5400*** -0.0601*** (0.0052) (0.0188) Return on Assets 0.0226*** 0.0659*** (0.0014) (0.0049) Capital ratio -0.7090*** 0.1350*** (0.0133) (0.0479) NPL -0.3710*** 5.9290*** (0.0366) (0.1320) Constant 0.9900*** -0.1680*** (0.0046) (0.0167) Observations 15,983 15,983 R-squared 0.602 0.157
Regression of risk-weighted assets to total assets (RWATA) and Charge-off ratio. The sample is generated from the FR Y-9C reports for the period 2013Q1 through 2017Q4. All variables are defined in Appendix A. Standard error in parentheses. *** p<0.01, ** p<0.05, * p<0.1.
