The relationship between credit supply and the strength of the CLO market

(1)

The relationship between credit supply and the

strength of the CLO market

University of Amsterdam, Amsterdam Business School

Msc Business Economics, Finance track

Master Thesis

Jacob Versteeg

6090923

Date: July 2, 2015

(2)

Abstract

This paper shows a positive relationship between the strength of the CLO market and credit supply provided by banks for medium- and large sized enterprises. This implies that banks originate more loans when they are able to distribute loans to third party investors. These results are robust to several estimation techniques, control variables and lagged effects. This research makes use of panel data that contains the size of the CLO market and the size of the commercial bank-loan market per quarter for seven European countries and for Europe as a whole. A time and entity fixed effects regression and instrumental variable regression model is used to control for omitted variable bias and to address potential simultaneous causality.

Statement of Originality

This document is written by Student Jacob Versteeg who declares to take full responsi-bility 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)

Glossary

ABCP Asset Backed Commercial Paper is commercial paper, a

promissory note with a fixed maturity of no more than 270 days, backed by an ABS

ABS An Asset Backed Security is both a security backed by

consumer loans and an umbrella term for all securitized products

Alt-A Alternative-A loans are riskier than A-loans but less risky than subprime loans

CDO Collateralized Debt Obligations were originally securities

backed by bank loans and bonds, but developed to a security that can be backed by all types of structured finance

products, like RMBS or other CDOs

CLO Collateralized Loan Obligations are a form of ABS where the

security is backed by commercial bank-loans

Commercial paper An unsecured promissory note with a fixed maturity of no more than 270 days

Equity tranche The equity tranche is the tranche that is served last in the waterfall payment structure of the ABS and is therefore the most risky tranche of an ABS (see appendix C)

Fannie Mae The Federal National Mortgage Association is a GSE whose purpose is to support the secondary mortgage market by securitizing mortgages in the form of mortgage-backed securities. Fannie Mae mainly buys loans from large commercial banks

Freddie Mac The Federal Home Loan Mortgage Corporation is similar to Fannie Mae, but buys their loans mainly from small banks Ginnie Mae The Government National Mortgage Association is a GSE

that promotes home ownership by providing affordable housing finance

GSE Government Sponsored Enterprises are financial service

corporations created by US Congress to enhance the credit flow of a targeted sector

(4)

Jumbo loans A loan with sufficient credit quality that is too large to quality for a GSE purchase

Maturity mismatch A bank’s assets and liabilities have a maturity mismatch if its assets have longer maturities than its liabilities. This is often the case since for example mortgages have longer maturities than consumer deposits on a cash account

MBS A Mortgage Backed Security is a security backed by

mortgages. The subgroups RMBS and CMBS are backed by residential and commercial mortgages respectively

Moral hazard Moral hazard is the phenomenon where one person takes more risk since another person bears the risk

Regulatory capital Regulatory capital is the amount of capital that a financial institution has to hold as determined by the regulator. According to the Basel rules, the regulatory capital is based on a bank’s assets, where each asset type has different weightings

Retained fraction The retained fraction of an ABS is the fraction that the issuing institution holds on its own balance sheet, rather than selling to third party investors

RMBS A Residential Mortgage Backed Security is a security backed

by residential mortgages

Securitization Securitization is the practice of converting illiquid loans held by financial institutions to liquid securities available for sale to third parties

SPV A Special Purpose Vehicle is a limited liability company that

buys assets in a securitization transaction and funds the asset purchase with a structured finance product

Sub-prime A loan to an individual who has difficulties maintaining the interest payment

(5)

1 Introduction

One key function of banks is to issue loans to companies, and banks are specialized in the screening and selection process of new loans. These loans are however illiquid and takes much regulatory capital. Securitization is the practice of converting illiquid loans held by financial institutions to liquid securities available for sale to third parties. By this transition, financial institutions are able to transfer risk to financial markets, reduce fund-ing costs and lower capital requirements. Financial institutions can for example originate and distribute receivables from mortgage-, credit card or car loans. This mechanism gives investors the opportunity to invest in assets without the need for a specialized screening process for individual loans. A structured and securitized portfolio of commercial bank-loans is called a Collateralized Loan Obligation (CLO). A more thorough background of CLOs is given in section 2.

In a strong CLO market with high demand for CLOs, financial institutions are easily able to originate and securitize loans. When the CLO market is weak, banks are not able to distribute loans and have to hold all originated loans on their balance sheets. Hence, the investor base is larger when this distribution mechanism is strong. This leads to the hypothesis that there is a positive relationship between the strength of the CLO market and the volume of commercial bank-loans.

The first modern securitized product was issued in 1970 by the US government-backed mortgage association Ginnie Mae. US congress passed a bill to resolve the concern that the existing system was not able to finance the large housing demand. Government Spon-sored Enterprises (GSE) such as Ginnie Mae and Freddie Mac were formed to promote home ownership. The GSEs use securitization in order to create a secondary market for mortgages. The size of the securitization market experienced a strong growth after the 1970s. The size of securitized mortgages was around $28 billion in 1976 and other securitized assets were still non-existent. According to Sifma statistics, the size of the US mortgage marked was around $9.4 trillion dollar in 2007. The global CLO market grew from $344 million in 1989 to $95 billion in 2000 and $526 billion in 2007. The total European market for Asset Backed Securities (ABS) grew to $7.4 trillion in 2007 of which $927 billion is consumer related, $4.6 trillion is mortgage related and $1.0 trillion consists of Collateralized Debt Obligations (CDO). The issuance of securitized products however faced an abrupt decline during the 2007-2008 credit crisis. The total value of European

(7)

new issues was $589 billion in 2009, compared to $1.2 trillion one year before.

There is a rich literature that explains the role of securitization during the 2007-2008 financial crisis, as summarized in section 2. Current literature focus on the relationship between securitization and mortgage credit growth, for several markets and mechanisms. For example, Mian and Sufi (2008) show that the supply shift of mortgage loans is driven by securitization, Rajan et al. (2010), Piskorski et al. (2010) and Parlour and Plantin (2008) analyze the relationship between secutitization and ex-post monitoring and Pur-nanandam (2011) shows the relationship between securitization and credit quality dete-rioration. The mortgage market differs however in kind from other securitizable assets since the US mortgage market is sponsored and actively managed by GSEs. Findings on the mortgage market therefore lacks external validity. There is still no literature available giving evidence about the relationship between the strength of the CLO market and the size of commercial bank-loans. European companies faced financing constraints during the economic crisis since banks lacked in their credit providing role. Credit provision is vital for investments and economic growth. This paper contributes to the evolving dis-cussion of the traditional role of banking and its relation with the real economy. The outcome of this paper is important on a macro level, as it gives insight in the transition of traditional bank roles and is important for regulatory purposes.

This paper tests the relationship between the strength of the CLO market and the volume of originated commercial bank-loans using a time and entity fixed effects regression model and an instrumental variable regression model. A unique data set is constructed using information from ConceptABS. ConceptABS is a news wire and database that gives information about each issued European asset backed security with all relevant characteristics. A panel data set is constructed with the size of CLO issues per quarter for eight European regions.

The paper will continue with a literature review in part 2, data description and sum-mary statistics in part 3, methodology in part 4, results in part 5 and the conclusion and final remarks in part 6.

(8)

2 Literature review

2.1 Institutional background

2.1.1 Securitization

Securitization is the practice of converting illiquid loans held by financial institutions to liquid securities available for sale to third parties. The issuing institution pools and structures loans or other assets that generates receivables in order to distribute the struc-tured product to outside investors. The general name of a securitized product is an Asset Backed Security (ABS), which is confusingly both an umbrella term for all types of Asset Backed Securities and a term for securities backed by consumer loans. Appendix B gives an overview of some key types of ABS. The main types of ABS are Asset Backed Securities backed by consumer loans, Mortgage Backed Securities (MBS) backed by mortgage loans and Collateralized Debt Obligations (CDO). CDOs were initially backed by commercial bank-loans, but has evolved over time. Modern CDOs could be backed by receivables from loans, bonds, structured finance products (SFCDO) or other CDOs (CDOn_{). A}

CDO backed by commercial bank-loans is now called a Collateralized Loan Obligation (CLO).

There are several motives for institutions to securitize assets, where the following three are most important. First, securitization could reduce funding costs. A car lease company is for example BBB-rated but has lease receivables which are AA-rated. When tapping debt markets, the company has to pay an interest level according to its BBB-rating. When the company however is able to isolate its AA-rated receivables and issue a loan with the AA-rated receivables as collateral, the company is able to pay an interest rate based on the AA-rated receivables. Second, banks provide loans to their customers and have to hold regulatory capital for each loan. Banks are originally the main investors in loans since banks are specialized in the screening and monitoring process. By securitizing originated loans, banks are able to sell these securitized loans to outside investors, which reduces the regulatory capital of banks. This cleared regulatory capital could be used to originate and distribute more loans in order to capture a fee, or could be used to change the asset mix of the bank. CLOs issued with this purpose are called balance sheet CLOs. Third, securitization is used for arbitrage purposes. The so called arbitrage CLOs are issued with the purpose of adding value by repackaging the underlying loans. The proceeds of the

(9)

issued CLO are used to buy a portfolio of loans which cash flows are pledged to make the regular interest and principal payments. The arbitrage is a result of the spread difference between the non-investment grade and illiquid bank-loans and the investment grade and liquid CLO. The greater the difference between the coupon payments of the underlying loans and the interest payments on the CLO, the higher the return on equity. This is however no true arbitrage since the bank often gives explicit guarantees to CLO investors (Acharya et al., 2013) and bears a risk due to a potential maturity mismatch.

2.1.2 Issuance

This paragraph describes the typical mechanics of an ABS issue. The financial institution that issues the ABS sets up a Special Purpose Vehicle (SPV) that buys the securitized assets. The SPV issues structured finance products to finance the transaction and pays the principal and interest with the receivables from the assets. The SPV is bankruptcy re-mote, meaning that creditors of the SPV are the only party with creditor rights. Although the SPV is a subsidiary of the financial institution that issues the ABS, creditors of that financial institution have no right to collect collateral, even if the parent company is in default. Appendix C summarizes the issuance process in a figure. The structured finance products, called asset backed securities, are often tranched with different risk character-istics per tranche. The proceeds of the receivables are distributed over the tranches like a waterfall. The proceeds will flow to the first tranche and follows with the second tranche when the first tranche is fully paid. This process continues until all proceeds are fully distributed. As a consequence, the first tranche is most safe and often rated AAA. The last tranche, called the equity tranche, is most risky and often retained by the issuing bank. Appendix D step 1 and 2 shows the waterfall structure of residential mortgage payments to Residential Mortgage Backed Securities (RMBS).

The issuer could in addition use a credit enhancer, rating agency, servicing agency and/or an underwriter. With a credit enhancer, the payments to the tranche are more or less guaranteed. This could be in the form of over-collateralization, third-party insurance or by holding a cash account. In the case of over-collateralization, the issuer posts more collateral than is needed so that the ABS is safer and has a better debt rating. The issuer could for example sell an AC80 million RMBS that is backed by mortgages that are worth AC100 million. In that case, investors only face losses when more than 20% of the mortgage

(10)

loans default. The servicing agent takes care of all cash flows from and to the SPV, the underwriter helps the SPV with the sale of the CLO and the rating agency provides a credit rating for the CLO.

2.1.3 Crisis

Securitization was one of the key drivers of the large mortgage growth and lowered mort-gage rates, as analyzed by Black et al. (1981), Heuson et al. (2001) and Kolari et al. (1998). However, the loose regulated securitization market was also one of the key drivers of the economic crisis of 2007 and 2008 (Acharya et al., 2013). This section gives a brief overview of the role of securization in the last crisis.

Mortgage providers securitized mortgages on a large scale and sold them to third party investors as Residential Mortgage Backed Securities (RMBS). Financial institutions cre-ated in their turn CDOs from the equity tranches of a portfolio of RMBS, as shown in panel 3 of appendix D. The equity tranches of CDOs could then be securitized into CDO2_.

These products were appealing to investors due to the high diversification and attractive returns. This securitization process functioned as a flywheel for mortgage growth, but also faced moral hazard issues. In the originate and distribute model, where mortgage providers originate loans to distribute them to third party investors, mortgage providers lack incentives to screen and monitor the originated loans. Hence, loans were originated with insufficient credit quality, which are called sub-prime or alt-A loans. These individ-uals were able to buy expensive houses for a teaser interest rate in the first years, and this system worked until housing prices started to decline. Sub-prime lenders defaulted on their payments which triggered the burst of the bubble. Mortgage providers disposed collateralized houses at discounts, which amplified the price decline. As a result, the super-safe perceived RMBS started to face troubles and many equity tranches became worthless. CDOs that were made of a portfolio of RMBS started to default and so did the CDOn_{s. Rating agencies started to downgrade these products. According to the Basel}

II banking regulation, banks have to hold regulatory capital based on the rating of each product. The RMBS and CDO defaults had therefore not only a direct effect, but also an indirect effect since banks had to hold more regulatory capital. This was the first cause of the liquidity problems for banks.

(11)

liabilities. This is profitable since long-term loans have higher interest rates, but this practice also faces liquidity risk when funding stops. This so called maturity mismatch is therefore strictly regulated by bank regulators. This regulation however does not apply to the shadow banking sector, which includes SVPs. In order to profit from the maturity mismatch, banks originated long term loans and financed the transaction with short term Asset Backed Commercial Paper (ABCP). This commercial paper, a promissory note with a maturity less than 270 days, was backed by securitized products like RMBS and CDOs. Acharya et al. (2013) show that this ABCP market grew to around $1.3 trillion in 2007, and that funding of banks rely significantly on ABCP. This profitable practice continued to be profitable until the ABS market collapsed. Financiers of the short-term ABCP stopped investing in the ABCP market so that the funding of commercial banks diminished. Data provided by the Federal Bank of St. Louis show that the ABCP market declined after its $1.3 trillion peak in 2007 to around $700 billion three months later. The assets that were funded with the ABCPs had longer maturities and could therefore not be unwinded. In order to maintain its capital ratio, banks should either sell assets or find alternative funding. Other financial institutions faced however the same problem and therefore funding and the demand for long-term assets were scarce. This caused an immediate liquidity squeeze at financial institutions. This modern version of a bank run is called the ABCP shadow banking run and this liquidity squeeze forced governments to bail out financial institutions.

2.2 Securitization and mortgage credit growth

Securitization affects the traditional view on deposit institutions and the nature of bank-ing. The traditional role of banks is that of a credit provider (Kashyap et al., 2002), while banks fulfill the role of an intermediary with securitization practices (Loutskina and Strahan, 2011). As an intermediary, banks could exploit securitization to provide credit without the need of a large balance sheet. This transition has however several implications.

Deregulation in financial markets and loan securitization have led to larger diversifica-tion. Loutskina and Strahan (2011) discuss the implications of geographic diversificadiversifica-tion. Due to deregulation, banks are able to operate around the globe and due to securitization banks are able to invest in foreign loans. This diversification should protect banks from

(12)

local shocks. The authors show that this diversification however comes at a price. The article compares investments made by concentrated lenders versus diversified lenders in the jumbo versus non-jumbo loan market. Jumbo loans are loans with sufficient credit quality, but too large to qualify for a GSE purchase. GSEs however subsidize non-jumbo loans since several GSEs stand ready to purchase the loans. This purchase guarantee eliminates the need to invest in private information. The authors show that concentrated lenders are more active in the jumbo market, and that these lenders therefore need to invest more in private information. The share of mortgage loans that are issued by con-centrated, non-diversified investors however fell from 18% in 1992 to around 4% in 2004. This shows that the compositional shift towards diversified lenders reduced information gathering by banks, and the authors indicate that this might have played a role in the development of the housing bubble. In addition, the authors first show empirically that concentrated lenders accept a higher portion of mortgage applications, which suggests that they are better able to price risk. Second, using panel regressions the article shows that concentrated lenders make higher returns, that their systemic risk is lower and their stock prices fell less than diversified lenders during the 2007-2008 crisis. To conclude, the authors show that diversification comes at a cost and that not all lending should be done by diversified lenders, as it was almost the case in the US mortgage market in the recent past.

Several studies show the relationship between mortgage growth and securitization. Mian and Sufi (2008) show that the supply shift of mortgage loans is driven by securi-tization, using a sample of 2002-2005 new home purchases. They show a skewed supply shift towards zip codes that contain a higher fraction of subprime borrowers. This impli-cates that the mortgage growth by securitization was driven by the growth of subprime mortgage growth. Demyanyk and Van Hemert (2011) provide evidence on the relation-ship between mortgage growth and securitization using loan-level data of subprime US mortgages. They found that the loan quality deteriorated together with the growth of mortgages and its securitized products. In addition, they show that securitization caused an unsustainable growth that led to a typical boom-bust cycle in the market. These problems could have been foreseen but were masked by the rising housing prices. Keys et al. (2008) show that this increase in mortgage supply after securitization reduces the screening incentives of financial intermediaries using a US dataset on subprime mortgage

(13)

loans. They exploit a rule of thumb that gives the likelihood that a loan portfolio is being securitized. The loan portfolio that is likely to be securitized is defaulting by a 10% to 25% higher chance than a loan portfolio with a similar risk profile but a lower probability of being securitized1_{. The paper does not make a comparison of default rates}

of non-securitized loan portfolios, so that the results are conditional on the portfolio being securitized.

Rajan et al. (2010), Piskorski et al. (2010) and Parlour and Plantin (2008) analyze the relationship between securitization and ex-post monitoring incentives and show that secu-ritization jeopardizes the fundamental screening and monitoring roles of banks. Piskorski et al. (2010) find a significant lower default rate for bank-held loans than securitized loans, conditional on the loan being delinquent and across various characteristics and time frames. The authors recognize several reasons why securitized loans face higher de-fault rates. First, banks may have different incentives to monitor bank-held loans and securitized loans, since the bank has a stronger monetary incentive on bank-held loans. Second, even when incentives are perfectly aligned, the pooling and servicing agreement may legally limit contract renegotiations by the servicer. Third, the dispersion of property rights creates a coordination problem among investors that hinder contract alterations. Fourth, lenders could delay foreclosures of portfolio mortgages and prioritize foreclosures of securitized loans for loss recognition purposes.

Purnanandam (2011) shows the relationship between securitization and credit quality deterioration. The paper concludes that banks that used originate-and-distribute models on a large scale during the pre-crisis era, originated excessively poor quality mortgage loans. This effect is larger for capital constrained banks. In addition, they show that a lack of screening incentives and risk-taking behavior that induces leverage significantly contributed to the crisis.

To conclude, current literature found evidence for the relationship between mortgage supply growth and securitization for several markets and mechanisms. The studies show that diversification due to securitization comes at a cost, that securitization weakens ex-post monitoring incentives and that securitization deteriorates credit quality in the mortgage market. These studies however lack external validity to other asset types, since the US mortgage market is unique in its kind. The US government actively fosters the

1_{A 25% higher chance means that the mean delinquency rate goes from for example 4% to 5%, not}

(14)

mortgage market with GSEs that encourage mortgage lending and home ownership. These GSEs, like Ginnie Mae, Fannie Mae and Freddie Mac guarantee the mortgage payments of confirming loans. These loans are originated by the GSEs and then securitized and issued to investors. The US mortgage market is therefore different in its kind than the corporate loan- and subsequent CLO market. The current literature lacks evidence on the relationship between corporate loan credit growth and CLO securitization. Hence, this paper contributes to the discussion about securitization and the transition of the traditional role of banking, since it shows how securitization affects the supply of corporate loans.

3 Data

3.1 Data sources

A unique database is constructed with data from ConceptABS. ConceptABS is a combined primary market news service and database dedicated to the European ABS market. The database contains all European ABS deals of each ABS type, each currency and all types of issuers. This paper focuses on CLO deals denominated in euros. The database contains deals as per January 2003 to April 2015. Hence, deals prior to the crisis, during the crisis and after the crisis are included. The database contains the announcement date, the deal name, SPV name, originator name, the type of issue, the collateral, the region and the lead investment banks per issue. All tranches per issue are also included with information about the size of the tranche, the credit rating, the credit enhancement, the weighted average life, the spread, the index to which the spread refers, and whether the tranche is marketed or retained. Deals that are fully retained are also included in the database.

There are in total 652 Euro-denominated CLO deals consisting of 3,638 tranches, is-sued by 649 unique Special Purpose Vehicles from 184 unique originators. Euro-denominated issues that include one or more non-euro currency tranches are excluded. Transactions that miss the country identifier, or whether the deal is marketed or retained are also excluded. After all exclusions, the data set consists of 586 deals.

ConceptABS gives the region and announcement date of each CLO, which gives the possibility to construct a panel data set. ConceptABS gives the sole country when the CLO is issued in a single country, the percentage per country if the CLO is formed from

(15)

loans from multiple countries, or gives ”Cross Boarder Europe” if the CLO is formed from loans from multiple countries and when these countries are not specified. In some cases, ConceptABS gives the region within the country, which is then replaced by the country. To construct the panel data set, the countries that have enough quarterly issues to contribute to the data set are determined. These are France, Germany, Italy, the Netherlands, Portugal, Spain and the UK. Second, for each issue, the relevant country or countries are determined. For issues that belong to several countries, the size of the issue is divided pro rata per country. For example, a AC100 million CLO that is formed from 40% German loans, 30% from loans from Valencia and 30% from loans from Catalonia, contributes for AC40 million to Germany and AC60 million to Spain.

Banks can obtain short-term funding from the ECB when they fulfill several require-ments. The ECB provides only funding when eligible collateral is deposited at the ECB2_.

The ECB does not accept individual loans since they are not able to assess each loan on an individual level. When these loans are however structured into a credit rated CLO, the ECB is able to assess the product on its eligibility. CLOs count as eligible collateral when they are AAA rated at issue and at least single A rated afterwards. The temporary frame-work introduced in July 2013 relaxes this condition to CLOs that are at least single-A rated at issue and afterwards. Due to the possibility to pledge CLOs as collateral, banks issued CLOs in order to receive ECB funding. These issues are recognized in the data set since these issues are fully retained. Fully retained issues have however no influence on the growth of bank loans. Although the ECB could be seen as a large buyer of CLOs, a fully retained CLO does not clear a bank’s balance sheet. In addition, it could be argued that banks use new issued loans as collateral and therefore fully retained CLOs influence credit growth. Banks that applied for ECB funding during the credit crisis were in need of capital and were therefore reluctant in issuing new loans. Therefore, banks securitize existing loans into fully retained CLOs. Hence, the argument that banks issue more loans after the securitization of a loan portfolio does only hold for marketed CLOs. To answer the question whether there is a relationship between CLOs and bank loan volume, only CLOs where at least one tranche is marketed are therefore included.

A variable is created that gives the size of each country’s CLO market per quarter, for issues with at least one marketed tranche. Each country’s CLO market size is linked to its

(16)

corresponding bank-loan market size. Deals that are specified as ”Cross Border Europe” are linked with the total European bank-loan size.

The bank-loan volume is available at the Statistical Data Warehouse from the Euro-pean Central Bank. Long- and short term loans from banks to non-financial medium and large sized corporations are obtained at a quarterly interval as per January 2003 to April 2015 for the six used Euro-zone countries and the total European loan size. The long-and short term bank-loan volumes to non-financial medium long-and large sized corporations in the UK are obtained from the website of the Bank of England at a quarterly interval. Each CLO issue has either a balance sheet or arbitrage purpose, and it is interesting to know whether the two groups have different effects on credit growth. Balance sheet CLOs are securitizations of corporate loans on a bank’s balance sheet, and have mainly two objectives. First, although originating corporate loans is important for a bank’s relationships, it takes much regulatory capital. Balance sheet CLOs could be used as a mechanism to lower this regulatory capital. The on-balance sheet loan portfolio is taken off-balance sheet by selling the portfolio to the bank’s SPV. The credit risk is thereby transferred to third party investors that invested in the CLO. Now the loan portfolio and credit risk is taken off-balance sheet, the balance sheet offers room to issue new loans. Second, balance sheet CLOs are used for funding purposes. Banks can issue either unsecured bonds, covered bonds or CLOs. Contrary to covered bonds, investors in CLOs are the only claimants on the collateral and the SPV is bankruptcy remote. Therefore, the default risk of a CLO is lower than the default risk of a covered bond so that the interest spread is lower. Since banks have typically large loan portfolios, they can securitize these loans for funding purposes. Arbitrage CLOs on the other hand, are CLOs issued with the purpose of adding value by repackaging the underlying loans, as described in section 2.1.1.

There are mainly three differences between the two CLO types. First, the main in-centive for a balance sheet CLO issue is to obtain regulatory capital relief and to lower funding costs, while arbitrage CLOs are used to make profits by securitizing illiquid and non-investment grade loans to a liquid and investment-grade CLO. Second, the depart-ment where the decisions are made differs between both types. Balance sheet CLOs are issued by the bank’s treasury, while arbitrage CLOs are issued by the investment banking division of the bank. The third key difference is the alignment of interest or moral hazard

(17)

problem. The issuing bank is often the only party with information about the underlying loans. This feature of CLOs could incentivise banks to structure bad loans to clear their balance sheet, as shown by Piskorski et al. (2010). Since balance sheet CLOs face these moral hazard problems and are accused of being a key cause of the credit crisis, it is interesting to see whether the two groups behave differently. ConceptABS gives for each issue whether the issue has an arbitrage or balance sheet purpose. A dummy variable is created that returns one in case of a balance sheet issue. Since the panel data set has a quarterly time frame, a variable is created that gives the mean of the dummy variable. This could be interpreted as the ratio of balance sheet CLO issues relative to all CLO issues per quarter and per country. In addition, the first two lags of CLO issue volume are constructed, so that a lagged effect in the relationship between CLO issue volume and bank-loan volume could be determined.

3.2 Summary statistics

Figure 1 shows the CLO market over time, where the solid line shows the size of marketed issues and the dotted line the size of the fully retained issues. The figure shows a collapse of the market in 2007, after its all time high in 2006. The size of marketed CLOs declined from around AC35 billion in 2007 to AC6 billion one year later. The market revives as per 2012, after a period where the CLO market has totally evaporated. The graph also shows the sharp increase in fully-retained CLO issues. This seems odd at first sight but makes actually perfect sense. Due to the crisis and the collapse of the ABCP market, banks were in immediate need of liquidity. The European Central Bank offers funding to banks and accepts CLOs as collateral. This ECB funding was extremely important for banks during the credit crisis in 2009 and the Euro crisis in 2011. Banks therefore issued CLOs and fully retained all tranches for ECB collateral purposes. This clarifies the strong increase in fully retained CLOs as per 2007 and the peaks in 2009 and 2011.

Table 1 panel A shows the number of CLO issues per year. There are in total 586 issues of which 272 issues are fully retained. The table shows that there were no fully retained deals prior to 2006. This makes sense since fully retained CLOs are issued for arbitrage purposes and pledged as collateral at the ECB. There were no marketed deals during the years 2010 to 2012. This also makes sense since CLOs were not competitive as funding instrument during the crisis years. Panel B shows the total, mean, minimum

(18)

and maximum issue sizes per year. The average CLO issue is AC1.1 billion, the average fully retained issue AC1.4 billion and the average marketed issueAC490 million. This shows that banks issue large CLOs to pledge as ECB collateral. The largest fully retained issue is AC10.8 billion. Panel A and B further shows that the peak in retained deals was during the 2008-2009 credit crisis. Panel C shows the distribution of balance sheet CLOs and arbitrage CLOs. The table shows that the fully retained issues are classified as arbitrage deals. This is due to the fact that fully retained CLOs do not clear regulatory capital. Fully retained issues could also not be used for funding purposes from funding from the market, due to the simple fact that no tranches are sold.

Table 2 shows the panel data, although the data as used in the regressions have quarterly intervals. The table shows that there were no marketed CLO issues in the period from 2010 to 2012. The table also shows that the only marketed CLO in 2009 was issued in Spain.

(19)

T able 1: Summary statistics P anel A: Distribution of issues p er y ear All 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 Issues 586 30 31 44 103 91 77 49 25 29 25 35 47 F raction 1 0.05 0.05 0.07 0.17 0.15 0.13 0.08 0.04 0.04 0.04 0.05 0.08 Mark eted 314 30 31 44 91 53 9 1 0 0 0 20 35 F ully retained 272 0 0 0 12 38 68 48 25 29 25 15 12 P anel B: Summary statistics Size (ACm) All 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 T otal 583,187 13,247 15,688 34,859 60,608 80,879 71,525 91,562 41,407 63,407 42,720 28,477 38,801 Mean 1,097 441 506 792 588 862 920 1,868 1,657 2,186 1,708 813 825 Min 12 76 12 39 145 60 75 135 39 275 50 300 196 Max 10,763 1,800 2,000 6,750 2,953 7,500 10,763 8,280 10,059 9,290 6,587 5,249 9,260 Mark eted 184,583 13,247 15,688 34,859 52,397 35,017 5,772 585 0 0 0 10,872 16,142 Mean 490 423 360 605 401 431 613 585 0 0 0 543 446 Min 2 23 6 2 18 4 39 585 0 0 0 300 308 Max 6,750 1,800 2,000 6,750 2,953 3,545 1,500 585 0 0 0 2,040 927 F ully retained 398,604 0 0 0 8,210 45,861 65,753 90,977 41,407 63,407 42,720 17,605 22,659 Mean 1,415 0 0 0 481 1,078 921 1,656 1,694 2,153 1,708 1,152 1,888 Min 2 0 0 0 23 31 2 37 26 60 50 51 196 Max 10,763 0 0 0 2,030 7,500 10,763 8,280 10, 059 9,290 6,587 5,249 9,260 P anel C: Distribution of issues b y issue typ e Size (ACm) All 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 Balance sheet 239,203 12,247 14,868 12,778 24,192 63,762 55,217 2,946 0 0 0 14,389 38,801 Arbitrage 343,983 1,000 820 22,081 36,416 17,116 16,308 88,616 41,407 63,407 42,720 14,087 0

(20)

T able 2: P anel distributions P anel A: Distribution of all issued CLOs Size (ACm) All 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 Europ e 171,605 4,244 1,536 2,957 22,821 23,550 23,019 38,628 11,270 8,294 10,711 6,523 13,916 F rance 15,130 371 1, 096 1,471 3,032 2,430 816 1,531 2,760 74 1,452 93 0 German y 49,751 2,135 1,776 9,134 8,354 8,079 6,634 12,207 0 492 0 936 0 Netherlands 56,737 36 372 7,315 2,232 14,205 10,780 1,444 10,059 1,029 0 0 9,260 Italy 52,879 0 645 123 230 213 481 3,381 0 14,358 18,086 10,335 5,023 P ortugal 21,380 0 204 500 1,472 0 3,094 0 5,928 6,800 0 834 1,437 Spain 208,238 6,252 9,242 11,522 18,190 29, 188 25,197 30,720 11,228 32,358 12,470 9,703 9,164 UK 15,700 208 812 1,834 4,273 3,211 1,500 3,649 159 0 0 51 0 P anel B: Distribution of mark eted CLOs Size (ACm) All 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 Europ e 67,971 4,244 1,536 2,957 20,120 11,649 3,723 0 0 0 0 6,002 13,606 F rance 7,759 371 1, 096 1,471 2,620 2,159 39 0 0 0 0 0 0 German y 30,748 2,135 1,776 9,134 7,407 7,485 1,956 0 0 0 0 852 0 Netherlands 10,488 36 372 7,315 2,038 724 0 0 0 0 0 0 0 Italy 3,409 0 645 123 174 213 0 0 0 0 0 1, 569 683 P ortugal 3,614 0 204 500 1,472 0 0 0 0 0 0 0 1,437 Spain 55,420 6,252 9,242 11,522 14,870 10, 084 0 585 0 0 0 2, 448 414 UK 9,300 208 812 1,834 3,692 2,700 51 0 0 0 0 0 0 P anel C: Distribution of fully retained CLOs Size (ACm) All 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 Europ e 103,633 0 0 0 2,700 11,901 19,295 38,628 11,270 8,294 10,711 520 309 F rance 7,371 0 0 0 412 270 776 1,531 2, 760 74 1,452 93 0 German y 19,003 0 0 0 947 594 4,677 12,207 0 492 0 84 0 Netherlands 46,248 0 0 0 193 13,481 10,780 1,444 10,059 1,029 0 0 9,260 Italy 49,470 0 0 0 56 0 481 3,381 0 14,358 18,086 8,766 4,339 P ortugal 16,657 0 0 0 0 0 3,094 0 5,928 6,800 0 834 0 Spain 149,818 0 0 0 3,319 19,103 25,197 30,135 11,228 32,358 12,470 7,255 8,750 UK 6,400 0 0 0 580 510 1,449 3,649 159 0 0 51 0

(21)

4 Methodology

4.1 Time and entity fixed effects regression

The relationship between CLO volume and commercial bank-loan volume is determined with a regression model. The constructed data set allows for a panel regression model that uses more information than OLS and addresses potential omitted variable bias. The panel data set-up with a fixed effects regression model controls for observed and unobserved variables that change over time but are constant per country and for variables that change per country but are constant over time. A variable that changes over time but is constant per country could be an interest rate change or a monetary policy decision from the ECB that affects bank-loan volumes. Variables that vary per country but does not change over time could be whether the country has an exchange- or bank driven credit market or whether the country use French or common law. It could also be the case that some variables change over time and per country so that a random effects regression model is most appropriate. Before constructing the model, it is important to perform several statistical tests. First I determine whether to include time fixed effects. Second, I show that a fixed effects regression model is most appropriate. Third, that heteroscedasticity, autocorrelation and serial correlation consistent standard errors should be used and fourth that the loan volume and CLO variables does not contain a unit root. All test outcomes are summarized in appendix A.

To test whether to include time fixed effects, a fixed effects panel regression model is constructed that includes dummy variables for each quarter. An F-test is performed that tests whether the dummy coefficients are different from zero. The test shows for a 5% significance level that the included time dummies contribute to the fit of the model.

A Breusch-Pagan Lagrange multiplier (B-P/LM) test is used to test if there is any variance across countries. The null-hypothesis is that the variance across countries is zero and therefore OLS or a time-series model is most appropriate. The alternative hypothesis gives evidence to use a fixed effects or random effects regression model. The B-P/LM test rejects the null-hypothesis for a significance level of 5% that there is no variance between countries and therefore a panel data regression model is used.

To decide on the design of the panel regression, a Hausman test is conducted to decide between a fixed effects and random effects model. The Hausman test tests whether the

(22)

unique errors are correlated with the regressors. Under the null-hypothesis, the random effects design is most appropriate, while the fixed effects method is best under the al-ternative. The chi-squared outcome is -0.05 and therefore fails to meet the asymptotic assumptions of the Hausman test. Schreiber (2008) shows that the Hausman chi-squared test can be negative under the alternative hypothesis, not only in small samples but even asymptotically. The author shows that the negative chi-squared outcome could only be compatible under the alternative hypothesis and should be interpreted accordingly. Hence, a time and entity fixed effects regression model is used.

The standard errors in a panel regression model need thorough analysis since the stan-dard errors are not only potentially heteroskedastic, but also potentially auto-correlated, cross-sectional dependent and serial correlated. To test for heteroskedasticity a modi-fied Wald test for groupwise heteroskedasticity is performed that is homoskedastic under the null-hypothesis. The null hypothesis is rejected for the 5% significance level, which concludes that the standard error treatment should be heteroskedastic robust.

According to Baltagi (2008), cross-sectional dependence could be a problem in large panels with long time-series. A Breusch-Pagan LM test of independence (B-P/LM) is used to test for cross-sectional dependence. The B-P/LM test gives that the residuals across entities are not correlated under the null hypothesis. The chi-squared result rejects the null-hypothesis for the 5% significance level, meaning that the handling of the standard errors in the regression model should be robust for cross-sectional dependence. In addi-tion, a Pesaran Cross-Sectional Dependence test is performed to test for cross sectional dependence, as suggested by Hoechle (2007). This test assesses whether the residuals are correlated across entities. The Pesaran test outcome gives evidence of across country correlated residuals for the 5% significance level. To address this cross-sectional depen-dence, Driscoll and Kraay (1998) present a covariance matrix estimation technique that yields standard error estimates that are robust to heteroskedasticity, serial correlation and general forms of cross-sectional dependence. Hence, the time and entity fixed effects re-gression model uses heteroskedastic, autocorrelated and serial correlated (HACSC) robust standard errors as suggested by Driscoll and Kraay (1998).

A unit root could cause problems in statistical inference and therefore a Levin–Lin–Chu (LLC) test is used to test for a unit root. The LLC test requires

q

N

T → 0, where N

(23)

The Harris–Tzavalis test requires N → 0 and the Breitung test require that T and N sequentially go to infinity. Since the panel is relatively small with N=8 and T=48, the LLC conditions are most appropriate. In addition, the LLC test is used in balanced panels in contrast to the Im–Pesaran–Shin and Hadri LM test. As the used panel is perfectly balanced, the LLC test is preferred. The performed LLC test uses an ADF regression with the number of lags chosen by the Akaike Information Criterion. The LLC unit root test rejects the null hypothesis that the panel contains a unit root for the 5% significance level for both the loan volume variable and the CLO variable.

Model 1 gives the time and entity fixed effects regression model. Model 2 gives the time and entity fixed effects regression model controlled for the time variable dummies. Model 3 and 4 are augmented models that control for the purpose of each issue. As discussed in the data section, balance sheet issues differ from arbitrage issues since balance sheet issues suffer from potential moral hazard problems. Hence, it is interesting to see whether the two groups behave different. The balance sheet variable is included in model 3 and 4, where model 4 includes time dummies. The balance sheet variable is the mean of a dummy variable, and should be interpreted as the ratio of balance sheet issues per country and per quarter. Model 5 and model 6 give two lagged values of CLO Volume and model 6 includes time dummies. The appropriate number of lags is determined with the Akaike Information Criterion.

In all specifications, ∆BLV stands for the change in bank-loan volume per quarter and per country, CLO Volume for the volume of issued CLOs per quarter and per country, Country Fixed Effects for all variables that are included due to the fixed effects regression design, Time Fixed Effects are the dummies for each quarter in each year and Balance Sheet is the ratio of balance sheet issues per country and per quarter.

∆BLVi,t = α + β × CLO V olumei,t+ Country F ixed Ef f ectsi + µi,t (1)

∆BLVi,t = α + β × CLO V olumei,t+ Country F ixed Ef f ectsi +

T ime F ixed Ef f ectst+ µi,t

(24)

∆BLVi,t = α + β × CLO V olumei,t+ γ × Balance Sheeti,t +

Country F ixed Ef f ectsi+ µi,t

(3)

Country F ixed Ef f ectsi+ T ime F ixed Ef f ectst+ µt

(4)

δ ×

2

X

i=1

CLO V olume : Si,t−n+ Country F ixed Ef f ectsi+ µi,t

(5)

δ ×

2

X

i=1

CLO V olume : Si,t−n+ Country F ixed Ef f ectsi +

T ime F ixed Ef f ectst+ µt

(6)

4.2 Instrumental variable regression

In addition to the fixed effects regression model, an instrumental variable (IV) regression model is used. The fixed effects regression design eliminates endogeneity by making use of panel data. The instrumental variable design also makes use of panel data. In addition, the endogenous variable is predicted by an exogenous variable that is not correlated with the error term of the explanatory model. The IV design also deals with potential simul-taneous causality. Hence, the IV regression design is better able to deal with potential endogeneity and eliminates potential simultaneous causality. In addition, the IV regres-sion contributes to the robustness of the results. This section introduces the instrument, test for the validity of the instrument and provides six specifications of the IV regression model.

One key reason for banks to issue CLOs is for funding purposes. Banks pursue a mixture of funding instruments, like unsecured bonds, covered bonds and securitized

(25)

products like CLOs. During normal times, unsecured bonds are most expensive due to its highest risk, followed by covered bonds and CLOs. Hence, CLOs are appealing as funding instruments during normal times. However, in August 2007 BNP Paribas announced to halt withdrawals from three CLO investment funds since the bank was not able to fairly value the funds after US subprime mortgage losses roiled credit markets. The credit crisis urged other banks to follow BNP Paribas and CLO interest spreads increased to levels far above unsecured and covered bond rates. CLOs lost thereby its advantage as a funding instrument and the number of marketed deals decreased. Therefore, there is a direct relationship between the interest spread of CLOs and the volume of marketed CLOs. There is however no relationship between the spreads of CLOs and the volumes of commercial bank-loans, other than via CLOs3.

The CLO spreads could be determined in the primary or secondary market. Since the spread at issue is determined based on peers that are traded in the secondary market, the two are highly interlinked. Secondary market prices were however at such high levels during the crisis years that is was difficult to price new issues with sufficient low spreads to produce attractive funding instruments. As a result, the primary CLO market collapsed so that there are not enough issues to determine the primary CLO interest spread. Hence, the CLO interest spread is determined in the secondary market. An index is created that consist of three closed-end CLO investment funds. These funds invest in a portfolio of CLOs and is therefore a good representation of the performance of CLOs. The index consists of the Pioneer floating rate fund, the Nuveen floating rate income fund and the Eaton Vance senior floating rate fund and is ranging from the first quarter of 2004 to the first quarter in 2015. As with bond pricing, the price of a CLO is the inverse of the interest rate, and therefore the return of the index decreases when spreads increases.

The first stage of the IV regression model regresses the index returns to the volume of marketed CLOs in order to get an exogenous predictor of the CLO variable. The instrument should be correlated with the endogenous explanatory variable in order to get predicted values, but may not be correlated with the error term in the explanatory equation in order to be valid. This paragraph shows the validity of the instrument.

The first stage regression gives a positive relationship between the index returns and

3_{The panel data design controls for omitted variable bias. Although there is no direct relationship}

between CLO spreads and the volume of commercial bank-loans, one could argue that the credit crisis caused CLO spreads to increase and bank-loan volumes to decrease. The time and entity fixed effects design controls for these effects.

(26)

the volume of marketed CLOs, implying a negative relationship between the spread and volume of marketed CLOs. With a standard error of 34.41 and a t-statistic of 4.26, the coefficient is significant for the 5% significance level. The standard error is determined using a two-way cluster with the region and time variable as clusters and a Bartlett kernel bandwidth of four. The bandwidth is determined using the rule of thumb that gives a bandwidth equal to T(1/4)_{. This is approximately two for T=44, but since the Bartlett}

kernel gives a weight of zero to the last lag a bandwidth of four is used. These statis-tics give evidence that the instrument is a strong predictor of the endogenous variable. Second, in order to test for strong identification, several tests are conducted. First, a Cragg-Donald Wald F test is conducted that assumes all variables to be independent and identically distributed (i.i.d.). With an F-statistic of 47.54, the null-hypothesis that the equation is weakly identified is rejected. The Kleibergen-Paap Wald rk F statistic does not need the assumption that the variables are i.i.d., and allows for heteroskedasticity and autocorrelated (HAC) standard errors. With an F-statistic of 18.14, the null-hypothesis that the equation is weakly identified is rejected. Both tests provide evidence that the instrument is strong. Third, to test for underidentification, a Kleibergen-Paap rk LM test is performed, giving a Lagrange Multiplier statistic of 5.867. This returns a Chi-squared p-value of 0.0154, which concludes that the model is identified for the 5% significance level. These tests give evidence that the model is identified and that the identification is strong and valid.

Models 7, 8 and 9 give the specifications of the instrumental variable regression models. Both stages are shown in separate lines. In specification 7 CLO V olumed _i,t is predicted by the spread in the secondary market, as shown in the first stage. In the second stage,

d

CLO V olumei,tis regresed on ∆BLVi,t, which is the change in bank loan volume per region

and quarter. Specifications 8 controls for the balance sheet variable and specification 9 controls for the first lagged value of the CLO variables. The appropriate number of lags is determined with the Akaike Information Criterion. Each specification, has a separate model that includes time fixed effects.

(27)

1st : CLO V olumed _i,t = α + β × Spread_t+ Country F ixed Ef f ects_i + µ_1,i,t 2nd : ∆BLVi,t = α + β ×CLO V olumed _i,t + Country F ixed Ef f ects_i +

µ2,i,t

(7)

1st : CLO V olumed _i,t = α + β × Spread_t + γ × Balance Sheet_i,t + Country F ixed Ef f ectsi + µ1,i,t

2nd : ∆BLVi,t = α + β ×CLO V olumed _i,t + γ × Balance Sheet_i,t + Country F ixed Ef f ectsi + µ2,i,t

(8)

1st : CLO V olumed _i,t = α + β × Spread_t + γ × Balance Sheet_i,t + δ × CLO V olume : Si,t−1 + Country F ixed

Ef f ectsi + µ1,i,t

2nd : ∆BLVi,t = α + β ×CLO V olumed _i,t + γ × Balance Sheet_i,t + δ × CLO V olume : Si,t−1 + Country F ixed

Ef f ectsi + µ2,i,t

(9)

5 Results

5.1 Fixed effects regression results

Table 3 gives the fixed effects regression results. All specifications give a positive and significant relationship between CLO volume and bank-loan volume. These results give evidence that support the hypothesis that there is a positive relationship between the strength of the CLO market and credit growth. In other words, banks tend to originate more loans when they are able to distribute the loans in the securitization market. Al-though CLOs have potential moral hazard issues and securitization contributed to the 2007-2008 credit crisis, these results show that a strong CLO market is important for credit supply.

(28)

for the 5% significance level. This means that the loan volume increases with 0.000523% after a AC1 million quarterly CLO issue in the average of the eight regions over time. This seems small, but knowing that the average loan level in February 2015 was AC603 billion, the coefficient is economically meaningful4_{. Specification 2 includes dummies that}

control for time fixed effects. The magnitude is now smaller than in specification 1, but still economically meaningful and significant for the 5% significance level. The changes in magnitude and significance indicate that the time dummies control for time fixed effects, as tested for in the methodology section. Also, the R-squared increases from 0.0825 to 0.5698, meaning that the statistical model in specification 2 gives a better fit.

In addition to the main hypothesis, table 3 shows whether balance sheet CLOs and arbitrage CLOs have different effects on credit supply. Specifications 3 and 4 give aug-mented models that include the balance sheet variable. The balance sheet variable is the mean of a dummy variable per quarter and per region, where the dummy variable equals one if the CLO issue has a balance sheet purpose and zero otherwise. The mean of this dummy is therefore the ratio of balance sheet CLOs relative to all CLOs. Specifications 3 and 4 give results that are not significant. This means that there is no difference between balance sheet CLOs and arbitrage CLOs regarding the relationship between the CLO market and credit growth.

Specifications 5 and 6 include two lagged values of CLO volume, that are used to determine whether there is a lagged effect. Specifications 5 and 6 give evidence that there is both a direct effect and a lagged effect, although the second lag in specification 6 is only significant for the 10% significance level. In both specifications, the number of observations decreases with 16, which is equal to two periods in eight regions.

5.2 Instrumental variable regression results

Table 4 gives the results from the IV regression model. Specification 1 gives a positive relationship with a coefficient of 0.0000256, significant for the 5% significance level. This should be interpreted as an increase of 0.00256% in loan levels after a AC1 million increase in quarterly CLO issues. This result is in line with the fixed effects regression results given in table 3, as the results give both significant and positive coefficients. The effect

4_{A coefficient of 0.00000523 means that the loan level increases with} _A_{C3.2 million after a}_A_{C1 million}

CLO increase. UK loan levels are converted to Euros using an exchange rate of EUR/GBP=0.70850, which is the official ECB exchange rate at 29 June 2015.

(29)

is however stronger in the IV model with a factor 5.

Specification 2 includes time fixed effects. The relationship is still positive and sig-nificant for the 5% significance level. The inclusion of time fixed effects however result in a less significant coefficient. This indicates that the time dummies control for omitted variables that are constant per region but vary over time.

Specifications 3 and 4 show whether the ratio of balance sheet CLOs is of importance for the relationship between the CLO market and credit growth. Specification 3 gives a relationship that is stronger than the relationship without inclusion of the balance sheet variable. The balance sheet variable itself is negative and significant for the 5% level. The coefficient of -0.147 should therefore be interpreted as a declining loan level of 0.000147% after a 10% increase in the balance sheet ratio. Since the balance sheet variable is a ratio, these results does however not give a negative relationship between balance sheet CLOs and credit growth. The results however does say that the relationship between the strength of the CLO market and credit growth is stronger for arbitrage CLOs. This relationship becomes however insignificant after inclusion of time fixed effects. This means that the negative relationship in specification 3 is driven by omitted variable bias. Hence, the results cannot conclude for the 5% significance level that the purpose of CLO issues is of importance for the relationship between the strength of the CLO market and credit growth.

Specifications 5 and 6 test for a potential lagged effect. The table clearly shows that the lagged coefficient is not significant, for both the specification with and without time fixed effects. These results are in contradiction with the fixed effects regression results given in table 3. Hence, the results as presented in table 3 and 4 cannot provide robust evidence of a lagged effect. The direct effect remains however significant for the 5% significance level. The number of observations in specification 5 and 6 in table 4 is equal to the number of observations in the specifications without the lagged value of CLO volume, in contrast to table 3. Data of the instrument is available as per 2004, while the CLO data is available as per 2003. Hence, the number of observations is determined by the instrument and not by the CLO volume variable.

(30)

T able 3: Fixed Effects Regression results The table rep orts the fixed effects regression results where the dep enden t v ariable is the change in total size of commercial bank-loan s. CLO V olume stands for the size of CLOs issued p er coun try and p er quarter. The balance sheet v ariable is the mean of a dumm y v ariable that returns 1 if the issue has a balance sheet purp ose and zero otherwise. Hence, the v ariable giv es the ratio of balance sheet issues p er quarter. CLO V olume: Si,t-n giv es tw o lagged v alues of CLO V olume. All sp ecifications use heteroscedasticit y , auto correlation and serial c or re lation consisten t standard errors. Sp ecifications 2, 3 and 5 include time fixed effects dummies. (1) (2) (3) (4) (5) (6) Loan change Loan change Loan change Loan change Loan change Loan change CLO V olume 0.00000523 ∗ 0.00000253 ∗ 0.00000501 ∗ 0.00000279 ∗ 0.00000265 ∗ 0.00000236 ∗ (0.000) (0.001) (0.000) (0.002) (0.003) (0.004) Balance sheet 0.00657 -0.00639 -0.00149 -0.00972 (0.743) (0.531) (0.933) (0.280) CLO V olume: Si,t-1 0.00000355 ∗ 0.000000976 ∗ (0.000) (0.046) CLO V olume: Si,t-2 0.00000480 ∗ 0.00000158 ∗∗ (0.000) (0.084) Constan t 1.005 ∗ 0.982 ∗ 1.005 ∗ 0.982 ∗ 1.003 ∗ 1.022 ∗ (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Observ ations 376 376 376 376 360 360 Within R 2 0.0825 0.5698 0.0834 0.5705 0.1950 0.5821 Time Fixed Effects Dummies No Y es No Y es No Y es p -v alues in paren theses ∗∗ p < 0 .10, ∗ p < 0 .05

(31)

T able 4: Instrumen tal V ariable Regression results The table rep orts the instrumen tal v ariable regression results. In the first stage, CLO V olume is predicted b y the in terest spread in the secondar y mark et p er quarter and p er region. In the second stage, the dep enden t v ariable is the change in total size of commercial bank-loans. CLO V olume stands for the estimator of CLO V olume as predicted in the first stage. The b alance sheet v ariable is the mean of a dumm y v ariable that return s 1 if the issue has a balance sheet purp ose and zero otherwise. Hence, th e v ariable giv es the ratio of balance sheet issues p er quarter. CLO V olume: Si,t-1 giv es the first lagged v alue of CLO V olume. All sp ecifications use heteroscedasticit y, auto corr e lati on and serial correlation consisten t standard errors. Sp ecifications 2, 3 and 5 include time fixed effects dummies. (1) (2) (3) (4) (5) (6) Loan change Loan change Loan change Loan change Loan change Loan change CLO V olume 0.0000256 ∗ 0.0000123 ∗ 0.0000311 ∗ 0.0000168 ∗ 0.0000386 ∗ 0.0000208 ∗ (0.000) (0.047) (0.001) (0.031) (0.006) (0.050) Balance sheet -0.147 ∗ -0.0702 -0.142 ∗ -0.0683 ∗∗ (0.004) (0.109) (0.005) (0.095) CLO V olume: Si,t-1 -0.00000589 -0.00000289 (0.224) (0.394) Constan t 0.995 ∗ 1.173 ∗ 0.999 ∗ 1.151 ∗ 0.999 ∗ 1.163 ∗ (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Observ ations 328 328 328 328 328 328 Time Fixed Effects No Y es No Y es No Y es p -v alues in paren theses ∗∗ p < 0 .10, ∗ p < 0 .05

(32)

6 Conclusion

This paper empirically shows the relationship between the strength of the CLO market and the volume of commercial bank loans. Current literature focuses on how several characteristics of securitization played a role during the 2007-2008 crisis. Other research finds the relationship between MBS and the size of the underlying mortgage market in the United States. These findings however lack external validity, not only since mortgage loans differ in kind from commercial bank-loans, but also since GSEs actively foster the US mortgage market.

A unique database is constructed that gives the size of issued CLOs and commercial bank-loans per quarter for seven European countries and for Europe as a whole between 2003 and 2014. In addition, the database contains an identifier that gives the purpose of each CLO issue. A time and entity fixed effects regression model is used that con-trols for omitted variable bias by making use of panel data. For robustness purposes, to further control for omitted variable bias and to address potential simultaneous causality, an instrumental variable regression is used that exploits the interest spread in the sec-ondary market as an instrument. The validity and strength of the instrument is shown empirically.

This paper shows a positive relationship between CLO volume and the change in bank loan volume, robust for several estimation techniques, control variables and lagged effects. These results show that banks originate more loans when the CLO market is strong. These results are in line with results found in the US mortgage market. In addition, this paper tests if balance sheet CLOs and arbitrage CLOs have different effects on credit growth. The results do not show robust evidence that the relationship differs between balance sheet CLOs and arbitrage CLOs. One of the limitations of the used approach is that the used data set-up does not allow for conclusions about the individual relationship between balance sheet CLOs and the credit market. In future work, it should be interesting to create separate balance sheet and arbitrage variables to test the absolute effect of the two CLO types.

(33)

Appendices

A

Statistical tests

Panel A: Test for time fixed effects

F(40, 279) 3.14

Prob > F 0.0000

Panel B: Test for entity fixed effects

Chi-squared 5,417.84

Prob > Chi-squared 0.0000

Panel C: Hausman test for appropriate panel regression design

Fixed Effects -136.0881

Random Effects -135.277

Difference -0.8102

Chi-squared -0.05

Panel D: Modified Wald test for groupwise heteroskedasticity

Chi-squared 19,426.27

Panel E: Cross-sectional dependence Breusch-Pagan LM test

Chi-squared 272.1707

Pesaran Cross-Sectional Dependence test

Pesaran’s test outcome 13.795

Probability 0.0000

Panel F: Unit root test

Unadjusted t -8.6744

Adjusted t* -5.6689

(34)

B

ABS overview

(35)

D

ABS Waterfall

This appendix shows the mechanism of a RMBS and CDO. Each house stands for one mortgage, where the white houses are defaulted mortgages. As the figure shows, some tranches will default when some of the underlying mortgages default. The CDO in the lower panel is created out of the equity tranches of all five RMBS. With this practice, one could make a AAA rated product out of risky assets. The CDO starts to make losses when some equity tranches of underlying RMBS start to default.

(36)

Bibliography

Acharya, Viral V., Philipp Schnabl, and Gustavo Suarez, 2013, Securitization without risk transfer, Journal of Financial Economics 107, 515–536.

Baltagi, Badi, 2008, Econometric analysis of panel data, volume 1 (John Wiley & Sons). Black, Deborah G., Kenneth D. Garbade, and William L. Silber, 1981, The impact of the gnma pass through program on fha mortgage costs, The Journal of Finance 36, 457–469.

Demyanyk, Yuliya, and Otto Van Hemert, 2011, Understanding the subprime mortgage crisis, Review of Financial Studies 24, 1848–1880.

Driscoll, John C, and Aart C Kraay, 1998, Consistent covariance matrix estimation with spatially dependent panel data, Review of economics and statistics 80, 549–560. Heuson, Andrea, Wayne Passmore, and Roger Sparks, 2001, Credit scoring and mortgage

securitization: Implications for mortgage rates and credit availability, The Journal of Real Estate Finance and Economics 23, 337–363.

Hoechle, Daniel, 2007, Robust standard errors for panel regressions with cross-sectional dependence, Stata Journal 7, 281.

Kashyap, Anil K., Raghuram Rajan, and Jeremy C. Stein, 2002, Banks as liquidity providers: An explanation for the coexistence of lending and deposit taking, The Jour-nal of Finance 57, 33–73.

Keys, Benjamin, Tanmoy Mukherjee, Amit Seru, and Vikrant Vig, 2008, Securitization and screening: Evidence from subprime mortgage backed securities, Quarterly Journal of Economics 125.

Kolari, James W., Donald R. Fraser, and Ali Anari, 1998, The effects of securitization on mortgage market yields: A cointegration analysis, Real Estate Economics 26, 677–693. Loutskina, Elena, and Philip E. Strahan, 2011, Informed and uninformed investment in housing: The downside of diversification, Review of Financial Studies 24, 1447–1480.

(37)

Mian, Atif, and Amir Sufi, 2008, The consequences of mortgage credit expansion: Evi-dence from the 2007 mortgage default crisis, NBER working paper series .

Parlour, Christine A., and Guillaume Plantin, 2008, Loan sales and relationship banking, The Journal of Finance 63, 1291–1314.

Piskorski, Tomasz, Amit Seru, and Vikrant Vig, 2010, Securitization and distressed loan renegotiation: Evidence from the subprime mortgage crisis, Journal of Financial Eco-nomics 97, 369–397.

Purnanandam, Amiyatosh, 2011, Originate-to-distribute model and the subprime mort-gage crisis, Review of Financial Studies 24, 1881–1915.

Rajan, Uday, Amit Seru, and Vikrant Vig, 2010, The failure of models that predict failure: distance, incentives and defaults, Chicago GSB Research Paper .

Schreiber, Sven, 2008, The hausman test statistic can be negative even asymptotically, Jahrbücher für Nationalökonomie und Statistik 394–405.

The relationship between credit supply and the strength of the CLO market