On loan sales, borrower and lender conditions : evidence from the U.S.

(1)

ON LOAN SALES, BORROWER AND LENDER

CONDITIONS: EVIDENCE FROM THE U.S.

MASTER THESIS

Fabian Mohr

Student-ID: 11087331

Submitted to the

University of Amsterdam, Amsterdam Business School

For the degree of

Master of Science in Business Economics – Finance

Thesis supervisor: Dr. Tomislav Ladika

July 2016

(2)

STATEMENT OF ORIGINALITY

This document is written by Fabian Mohr 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

In recent years there has been a tremendous growth in the trading of syndicated loans in the secondary market. This thesis uses a unique dataset for syndicated loans over the period of 1996 until 2013 to empirically investigate two major issues, using loan renegotiation events as a window into changes in the lending syndicate and to identify loan sales. First, I research whether syndicated loans are influenced by the borrowers’ performance – at origination and during the lifetime of the loan. Second, I examine whether loan sales are also influenced by the lenders’ performance. I find that especially the borrower’s EBITDA margin as a proxy for firm performance is inversely related to loan sales. Moreover, I find evidence that financially constrained lenders, measured by a low equity and tier 1 capital ratio, are more likely to sell their loan shares.

(4)

1 INTRODUCTION

An important topic in finance is the trading behavior and asset sales of financial institutions (Abassi et al., 2015). One important type of bank assets are loans, especially syndicated loans, which are the largest source of capital for firms nowadays (Nigro et al., 2010). A syndicated loan is a loan given to a borrower from a syndicate of lenders. Teams of banks usually issue loans, but there is very little empirical evidence on whether they retain or sell loan shares afterwards. This thesis adds to existing literature about loan sales by connecting loan sales directly to borrowers and lenders’ performance. The main contribution is to provide a new way to measure loan sales, and also that previous papers don't focus on linking loan sales to borrower and lender performance

.

This paper is most importantly related to the emerging literature on the relevance of the U.S. syndicated loan market and loan sales in particular. With the development of the syndicated loan market, banks began to keep only a small portion of the loans they originated, selling the remainder to other institutional investors (Paligorova and Santos, 2015). Also Yago and McCarthy (2004) find that the secondary loan sales market is dominated by leveraged, risky loans and most of the loans are purchased by non-bank, institutional investors.

Taking the outcomes of the literature review about loan sales and loan renegotiation into account and connecting this with further findings on loan funding in general, I hypothesize that the performance of a borrower is inversely related to sales in the loan syndicate: the worse the borrower’s performance, the higher the sale rate. Thus, lenders are more likely to sell their loan shares when the borrower is doing worse. Among other reasons, I argue that lenders sell loans of badly performing borrowers in order to avoid costly monitoring and to mitigate regulatory taxes such as capital requirements (see e.g., Dahiya et al., 2003; Drucker and Puri, 2009; Pennacchi, 1988). Furthermore and based on the same reasoning, I hypothesize that financially constrained lenders are more likely to sell their loan shares.

The research is based on datasets from my supervisor, Dr. Tomislav Ladika. Both datasets cover the years from 1996 until 2013. The first dataset contains 18,523 unique loans, and 25,223 unique loan tranches and their corresponding characteristics. A second dataset consists of information about 2,500 U.S. loan renegotiations. The loan renegotiation events are used as a

(6)

window into changes in the lending syndicate and to identify loan sale events. A loan is classified as sold if a lender who is a member of the syndicate of the original loan, is not in the syndicate of the very same, but renegotiated, loan anymore. Further information about the borrowers and lenders is included from the Compustat database. The final renegotiation dataset consists of 1,236 loans and 1,655 loan tranches, where I am able to test for possible connections of loan sales to lender and borrower performance. Furthermore, as robustness tests another approach is used to generate a dataset without renegotiation events. This dataset uses rollover events to identify loan sales. The final rollover dataset contains 1,678 loans and 1,910 loan tranches.

My empirical analyses starts with univariate results and continues with multivariate tests that base controls on existing work on the determinants on loan sales. The found results based on the borrower performance are somehow ambiguous. Depending on the performance measurements used, I find support of my hypothesis that the performance is inversely related to loan sales or not. Remarkable is herewith the development of the EBITDA margin, which seems to be strongly linked to loan sales in a way that, borrowers with a negative change in the EBITDA margin experience significantly more loan sale events. Moreover, the hypothesis that financially constrained lenders are more likely to sell their loan shares is supported by the findings, as the results are highly statistically significant and show that lenders with a lower equity ratio and a lower tier 1 capital ratio sell their loan shares more likely.

The remainder of this thesis is structured as follows. Section 2 summarizes the existing literature surrounding this topic and results in the formulation of the hypotheses. Section 3 describes the data, sample and the variables as well as the methodology. Section 4 presents both univariate and multivariate results on the relation between borrowers and lenders performance and loan sales. Section 5 consists of robustness checks. Section 6 provides discussion and the last section concludes. In the following, the terms “amendment” and “renegotiation “are synonymously used.

2 LITERATURE REVIEW AND HYPOTHESES DEVELOPMENT

This section starts with a review of the existing literature. As this thesis is the first to study the effect of borrower and lender conditions on sale rates, using loan renegotiation as a window into changes in the lending syndicate, I split up my research into the main categories surrounding this

(7)

topic. After a definition of what syndicate lending is and a short overview of the U.S. syndicated loan market, I will summarize empirical studies on loan sales and loan renegotiation. Taking these studies into account, I derive three hypotheses.

2.1 LITERATURE REVIEW

2.1.1 Important features of syndicate lending

Dennis and Mullineaux (2000) define a syndicate loan as two or more lending institutions jointly agree on terms and conditions and give together a single loan to a borrower. Herewith, usually one bank acts as a lead arranger and is therefore responsible of coordinating and organizing the whole process. The tasks of this lender start with the preparation of an information memorandum, where general information about the borrower, the borrower’s financial conditions and the proposed loan structure (e.g., loan amount, fees, maturity) are shown. Based on this memorandum the lead bank is responsible for approaching other lenders and coordinating the process. However, the lead arranger itself usually finances a large portion of the loan.

As borrowers have an incentive to misreport their quality to lenders, both at loan origination and after having received loans, borrowers and lenders are not likely to by equally informed. Lenders try to limit the unfavorable effects of the borrower’s information advantage by including covenants in the loan contracts (Drucker and Puri, 2009). Syndicate loans have numerous covenants that provide loan-holders considerable control and insight into a borrower’s health (Nigro et al., 2010). According to Assender (2000) and Miller (2004) loan covenants usually include restrictions on current ratios, such as maximum leverage ratios. Other common financial covenants include that the borrower has to maintain minimum net worth and current ratios. The covenants might even reduce the need for loan screening and monitoring (Berlin and Loeys, 1988).

2.1.2 The U.S. syndicated loan market

The syndicated loan market bridges the private and public debt markets and is one of the largest and most flexible sources of capital. During the past few years, syndicated loans have increasingly become the financial choice of large and mid-sized U.S. corporate borrowers (Nigro et al., 2010).

(8)

According to Dahiya et al. (2003), the syndicated loan market can be divided into two broad categories. The first category is the primary, or syndicated loan market. In this market portions of a loan are placed with a certain number of banks, often in connection and as part of the origination process. The second category is the seasoned, or secondary loan sale market. In this market, banks subsequently sell parts of the loan. Particularly the secondary loan market has grown enormously in the recent years (Yago and McCarty, 2004). Drucker and Puri (2009) find that the U.S. secondary loan market volume reached USD 238.6 billion in 2006 from USD 8.0 billion in 1992, which equals a compound annual growth rate of 25%.

2.1.3 Loan Sales

With the development of the syndicated loan market, banks began to keep only a small portion of the loans they originated, selling the remainder to other institutional investors (Paligorova and Santos, 2015). According to Drucker and Puri (2009), banks are increasingly selling loans to other banks and non-bank financial institutions such as pension funds, hedge funds, mutual funds, private equity firms and asset managers. They show that over 60% of loans are traded within one month of loan origination and nearly 90% are sold within one year. Moreover, Drucker and Puri (2009) show that loans are not only traded close to origination, but for an extended period of time – more than 50% of sold loans trade more than two years after origination. Similarly, Yago and McCarthy (2004) find that, in contrast with typical loan syndications, the secondary loan sales market is dominated by leveraged, risky loans and most of the loans are purchased by non-bank, institutional investors.

But why do banks sell their loan shares? Researchers have come up with several reasons. Secondary loan sales give originating banks the chance to diversify part of their credit risk by selling loans to other participants. With that, loan sales allow banks to comply with risk-adequacy regulations and internal risk controls (see e.g., Drucker and Puri, 2009). Selling loans also allows banks to continue investing in profitable projects even when capital is constrained (see e.g., Pavel and Phillis, 1987; and Pennacchi, 1988). Irani and Meisenzahl (2015) exploit a U.S. credit register of syndicated loans to track the dynamics of loan syndicates after origination. They show that banks reliant on wholesale funding attempt to smooth out funding disruptions by selling their shares on secondary markets. In contrast, banks reliant on more stable sources of funding are more likely to purchase loans. Thus, they empirically find that certain characteristics

(9)

of banks, which are members of the loan syndicate, influence loan sales. Banks could also sell their loan shares due to negative private information about the borrower (Berndt and Gupta, 2009). The loan buyers are then likely to be disadvantaged due to the lack of private information. This could lead to moral hazard and adverse selection problems (Penacchi, 1988; Gorten and Pennacchi, 1995). If banks sell their loan shares and thus reduce the exposure to the risks of the loans, they also reduce their incentive to engage in costly screening and monitoring of the borrowers (Paligorova and Santos, 2015).

From a borrower’s perspective, there are positive and negative consequences of their loans being sold. Positive effects include lower cost of capital (Gupta et al., 2008), increased access to debt (Drucker and Puri, 2009), and information effects (Gande and Saunders, 2012). However, the empirical literature also supports the idea that loan sales have negative consequences for the borrowers. Dahiya et al. (2003) evaluate the effects of loan sales in the secondary market on both borrowers as well as bank’s stock returns. While they do not find that the sale of a loan has s significant impact on the banks’ own stock return, they find evidence that loan sales have a negative effect on borrower’s returns. They also find that 42% of the borrowers whose loans are sold file for bankruptcy within three years after the announcement of a loan sale. In line with these findings, Berndt and Gupta (2009) report a 9% under-performance of borrowers whose loans are sold over the three-year period following the initial loan sale. They conclude that banks are either originating and selling loans of lower quality borrowers based on unobservable private information (adverse selection), and/or loan sales lead to less bank monitoring that affects borrowers negatively (moral hazard).

Which loan characteristics influence syndicated loan sales? Loan sales are positively related to the size and maturity of the loan (see e.g., Irani and Meisenzahl, 2015). Loans with higher deal amounts and longer maturities are more likely to be sold on the secondary market. Drucker and Puri (2009) show that loans sold in the secondary market carry more restrictive covenants, precisely to mitigate the drop in a bank’s monitoring effort. Moreover, sold loans are more likely to be term loans than loans with credit lines. They argue, besides other reasons, that credit lines require more intense monitoring because the borrower has an incentive to access the credit line when it is performing poorly.

(10)

2.1.4 Loan Renegotiation

Despite rich theoretical predictions and the importance of renegotiation in contract theory, few empirical studies are available on this subject.

Renegotiation of syndicated loan contracts is common and is done through amendments of the original loan agreements. Roberts and Sufi (2009) were the first to quantify the significance of renegotiation. Using a sample of private credit agreements between U.S. publicly traded firms and financial institutions, Roberts and Sufi (2009) show that over 90% of long-term debt contracts are renegotiated prior to their stated maturity. They find that renegotiations result in large changes to the amount, maturity and the pricing of the contract. Moreover and interestingly, they find that ex ante firm characteristics (e.g., leverage, profitability and volatility) and contract design choices (e.g., syndicate size) bear no relation to the incidence of renegotiation. These findings imply that new information is causing renegotiation. Based on this outcome, Roberts (2014) shows, using data from SEC filings, that the typical bank loan is renegotiated five times, on every nine months. Furthermore, the results show that loans for which the initial terms are restrictive, possibly due to high information asymmetry, are more likely to be renegotiated more frequently.

Sufi (2005) empirically explores the syndicated loan market with an emphasis on how information asymmetry and renegotiation considerations influence syndicate structure and the choice of participant lenders. He finds, among other things, that when the borrower is more likely to need to renegotiate the loan agreement, lead arrangers add participants with very small portions of the loan in the syndicate. They argue that adding participants with small portions reduces the renegotiation surplus expected by the borrower. Paligarova and Santos (2015) document that loans financed relatively more by non-bank lenders are associated with lower likelihood of renegotiation, while strong lead bank presence facilitates renegotiations.

2.2 HYPOTHESES DEVELOPMENT

Despite the importance of syndicated loans in corporate finance, research on the role of syndicated loans in U.S. corporate finance is limited. As shown in the literature overview, it has not been researched yet if banks are more likely to sell loans when a borrower or lender is doing worse. This thesis uses loan renegotiation events as a window into changes in the lending

(11)

syndicate, and then examines how loan sales are related to borrower and lender performance. On top of that, the created dataset is also used to test for possible connections of loan sales to lender performance.

Taking the outcomes of the literature review about loan sales and loan renegotiation into account and connecting this with further findings on loan funding in general, I hypothesize that the performance of a borrower is inversely related to sales in the loan syndicate: the worse the borrower’s performance, the higher the sale rate. Thus, banks are more likely to sell their loan shares when the borrower is doing worse. Therefore, two related borrower performance hypotheses, based on the same argumentation, are derived.

Hypothesis I: The performance of a borrower is inversely related to sales in the loan syndicate. Several arguments lead to this hypothesis. First, Yago and McCarthy (2004) find that, in contrast with typical loan syndications, the secondary loan sales market is dominated by leveraged, risky loans and most of the loans are purchased by non-bank, institutional investors. They find that the secondary market is dominated by badly performing borrowers with sub-investment grade ratings (Standard & Poor’s rating scale: BB or lower), in comparison to loans in the primary market. One reason for this might be that lenders participate in the loan structuring in order to receive high structuring fees at the very beginning, but try to sell their shares afterwards. When selling their shares, they are more likely to focus on the risky and badly performing borrowers. Second, Dahiya et al. (2003) find that 42% of the borrowers whose bilateral loans are sold file for bankruptcy within three years after the announcement of a loan sale. In line with these findings, Berndt and Gupta (2009) report a 9% under-performance of borrowers whose loans are sold over the three-year period following the initial loan sale. All authors conclude that banks are either originating and selling loans of lower quality borrowers based on unobservable private information (adverse selection), and/or loan sales lead to less bank monitoring that affects borrowers negatively (moral hazard). I argue that unobservable private information for badly performing borrowers is more valuable. The reasoning behind this is as follows: private information about the exact cash inflow, e.g. EBITDA, of a well performing borrower (high EBITDA) who has a cash inflow way above their financing costs anyway, is not so valuable from a lenders perspective as information about the exact cash inflow of a badly performing

(12)

borrower (low EBITDA). If a borrower has only an EBITDA that makes it difficult to pay its financing obligations, private information, e.g. high future losses, are valuable information for lenders. Consequently, I argue that lender, which are members of the loan syndicate from the origination on exploit their superior information by selling loans to more uninformed investors. Third, bank’s monitoring costs are increasing with the riskiness of the borrower’s business and the performance of the borrower itself (e.g., Berndt and Gupta, 2009). The worse the borrower performs, the more time and money the lender has to invest. In order to avoid those costs, lenders sell their loan shares.

Fourth, and another aspect to take into account is that banks have a desire to mitigate regulatory taxes such as capital requirements (see e.g., Dahiya et al., 2003; Drucker and Puri, 2009; Pennacchi, 1988). One capital requirement is the concept of risk-weighted assets (see e.g., Bruno et al., 2014), wherein banks are required by the supervisor to hold a minimum loss-absorbing capital to allow them to sustain through difficult markets. Depending how risky an asset is, they have to reserve more equity for that. Banks are required to hold more equity available for high risky loans in their portfolio. Since equity is a costly source of capital, I argue that banks therefore try to sell these loans.

Lastly, I argue, in line with Drucker and Puri (2009) and Bomfim (2005), that lenders can directly fund loans with the use of credit default swaps. However, this is mostly only possible for investment-grade loans. So, for loans not rated investment-grade, lenders may experience funding problems that can be solved with loan sales. Thus, I expect that the performance of a borrower, in this case measured by its rating, is inversely related to sales in the loan syndicate. Based on the mentioned explanations, I also derive my second borrower performance hypothesis.

Hypothesis II: The change in the borrower’s performance between origination and amendment is inversely related to sales in the loan syndicate.

As the first hypothesis is only looking at static variables at origination, I am now taking dynamic variables into account. Basically, the argumentation of the second hypothesis is the same as for the first hypothesis. However, the research on the second hypothesis now focuses more on the change of the borrower’s performance during the lifetime of the syndicated loan. In the case of

(13)

worsening borrower performance, loan sales might be a possibility for lenders to control their portfolio and credit risk (Drucker and Puri, 2009).

As of now, I focused on the borrower’s performance as a reason that influences loan sales. However, it might also be possible that loan sale in a syndicate is related to the lender’s performance.

Hypothesis III: Financially constrained lenders are more likely to sell their loan shares.

As mentioned before, lenders and especially banks have a desire to mitigate regulatory taxes such as capital requirements. In the aftermath of the financial crisis the supervisor increased their required capital demand even more. The more risky assets a lender holds, the more capital it has to have. Healthy and well performing lenders do not have a problem in providing the capital buffer. However, financially constraint banks might have issues with keeping the buffer at the required level. This argumentation is in line with Drucker and Puri (2009) who argue that loan sales allow banks to comply with risk-adequacy regulations and internal risk controls. Financially constrained lenders could therefore use loan sales in order to manage the utilization of their capital buffer.

This hypothesis is also connected to findings of Irani and Meisenzahl (2015). They show that banks reliant on wholesale funding attempt to smooth out funding disruptions by selling their shares on secondary markets. In contrast, banks reliant on more stable sources of funding are more likely to purchase loans. Thus, they empirically find that certain characteristics of banks, which are members of the loan syndicate, influence loan sales.

3 DATA AND METHODOLOGY

This section starts with a description of the time-consuming process of creating the databases that are used in this thesis. It also contains precise definitions of all variables and summary statistics for the complete dataset. In addition, this section also discusses the methodology that is used to test the three stated hypotheses.

(14)

3.1 DATA

3.1.1 Data sources and preparation of the amendment dataset

Main reason for the little empirical evidence so far is the limited data on loan renegotiation and loan sales available. Dr. Ladika provides me with a dataset containing 18,523 unique loans, and 25,223 unique loan tranches and their corresponding characteristics (e.g., name and loan amount of syndicate members, borrower identifier, maturity, markup, commitment) covering the years between 1996 and 2013. A loan identifier number uniquely identifies the loan tranches. Furthermore, he provides me with a dataset of over 2,500 U.S. loan renegotiations between 1996 and 2013. This datasets links an original loan to its responding renegotiated loan and contains the same loan identifier numbers as the dataset mentioned above. Based on the same loan identifier numbers it is therewith possible to link the two main datasets. To answer my research question, I just use the newly created dataset as a window to look at loan sales, hence the information about the lending syndicate of the original loan and the lending syndicate of the renegotiated loan is of utmost importance for me.

However, a unique identifier for lenders was mostly missing or incomplete in the dataset. Lenders can just be identified and distinguished between each other by their names. Unfortunately, the lender names where spelled and abbreviated differently in the dataset, which makes it very complicated to differentiate. According to Avraham et al. (2012) banking organizations in the U.S. are mostly organized in a bank holding structure, where bank subsidiaries engage in lending, deposit-taking and other activities. This makes it difficult to measure loan sales, since it might be the case that a loan share is just transferred internally to another subsidiary without actually selling the loan. It is therefore necessary to identify all branch offices and subsidiaries with different names, which all belong to the very same bank holding company. 1

Subsequently, I downloaded all financial companies (i.e. companies with SIC codes between

1_{Thus, with an extensive Internet search on the bank’s websites, I hand-match all lenders belonging to the same}

bank holding company and assign them a unique lender identifier (i.e. sdc_id) lasting from 1 to 838. In a next step, banks and non-banks were identified. Therefore, mainly the bank’s websites were used. If not available, other publicly available information (e.g. rating data or news articles) was used. Non-banks were further clustered into different groups – asset managers, collaterized funds and others. I tried to do this mostly by an extensive Internet search as well, however in doubt, the acronym at the name’s end was used. With LLC or LP in the end, the lenders are assigned as asset managers. The acronym CDO, CLO or other abbreviations are interpreted as collateralized funds. In cases with no match at all, those entities got assigned to “others”.

(15)

6000 and 6999) from both datasets (Compustat: North America and Global) and hand-matched this dataset with the main dataset described above on the basis of the name (Compustat item allmanagerslong). The aim here was to link the hand-created lender identifier to Compustat’s gvkey, so that I am able to include data on the lenders2. I end up with 755 matches. Based on the existing gvkey identifier, information about the borrowers is downloaded from the Compustat database.

To sum up, the result of the main data preparation is now a dataset, which contains original and the responding renegotiated loans with all their characteristics. The dataset consists of 1,236 loans, or, to be more specific, of 1,655 loan tranches. Moreover, it is now possible to include further data about lenders and borrowers on the basis of the created unique identifiers and their responding gvkeys. All variable definitions are summarized in Table I.

[Table I approximately here]

3.1.2 Main dependent variable: Loan sales

To measure loan sales, I create two loan sale variables out of the datasets described above. A syndicated loan often times consists of more tranches belonging to the same loan. Those tranches, which have different purposes and characteristics, are arranged under the same loan contract. The first tranche – typically called “A” – is mostly a redemption tranche with the shortest maturity, whereas the other tranches – typically called beginning with “B” – often times have a bullet structure with a longer maturity (see e.g., Armstrong, 2003; Dennis and Mullineaux, 2000). Due to the different characteristics, which might influence loan sales, I measure loan sales on the loan tranche level for my thesis. Therefore, I classify a loan as sold if a lender (identified by the created lender identifier variable), who is a member of the syndicate of the original loan, is not in the syndicate of the very same, but renegotiated/rolled over, loan anymore.

2_{In the first matching round I match all completely identical names and names which only differ in the use of capital}

or small letters or some abbreviations. After that, I also link reasonable names (e.g., I assume Alaska Pacific Bancorp to be the same as Alaska Pacific Bancshares Inc. or Associated Bank NA = Associated Banc-Corp). Thus, I found 341 matches (total lender names: 1,367), so 1,026 are still unmatched. Nevertheless, when I assign the found matches to all lenders belonging to one sdc_id (although obviously not the same, but at least belonging to the same entity).

(16)

My first variable of interest is therefore a binary variable “sale” taking the value one if a lender is not in the syndicate of the renegotiated/rolled over loan anymore and zero otherwise. With this approach, I have one observation for every borrower-loan tranche-lender combination, which allows me to include needed data on borrowers, loan characteristics and lenders. However, this also means I would overweight borrowers with a large syndicate as they would be disproportionately high present in the sample. Therefore, I also create a second dependent variable, “sale rate”. Sale rate is then the number of loan sales per loan tranche divided by the total amount of lenders in the syndicate at the origination date. Hence, sale rate is a continuous variable between zero – if no lender sold its loan tranche and quit the syndicate – and one – if all lenders sold their respective loan shares. The advantage of using this variable as the dependent variable is that there is only one observation for each borrower-loan tranche combination and consequentially borrowers with a large syndicate are not overweighed. However, with this approach it is not possible to include lender characteristics.

3.1.3 Performance measurements 3.1.3.1 Borrower

The following subsection discusses six performance measurements, both accounting-based and stock-based proxies that I use to evaluate the performance of a borrower in a loan syndicate. To link loan sales to borrower’s conditions, I have to evaluate the borrower’s performance. According to Snow and Hrebiniak (1980), there is no measurement that allows measuring every aspect of firm performance. However, accounting-based measures like Return-on-Assets (ROA), EBITDA margin and others are generally considered as an effective indicator to measure firm performance (see e.g., Al-Matari and Al-Swidi, 2014).

As a first measurement of firm performance, I therefore decide to make use of ROA, calculated as the ratio of net income (Compustat item NI) to total assets (Compustat item at).

Second and as an alternative, I use the EBITDA margin as a proxy for performance. EBITDA is earning before interest, taxes, depreciation and amortization and therefore measures the operating performance of the company excluding items, which are not directly related to the company’s core business. The smaller a company’s operating expenses in relation to total revenue, the higher this margin, indicating a more profitable operation. It also allows comparing

(17)

companies with different sizes or operating in different industries, since it brings down the company performance to a fraction of revenues. The margin is calculated as the ratio of EBITDA (Compustat item ebitda) to total revenues (Compustat item revt).

Third, a widely known measure focusing more on the firm’s bankruptcy risk is Altman’s Z Score (see e.g., Altman, 1968; Altman, 1977). Altman’s Z score predicts, based on accounting measures, the probability of the firm’s bankruptcy in the upcoming two years and could therefore also be used to test for firm performance.

Fourth, rating systems nowadays used are oftentimes still based on Altman’s Z-Score (see e.g., Miller, 2004). Due to information asymmetry credit rating got more very important in the credit process (see e.g., Hilscher and Wilson, 2015; Hung et al., 2016). Consequently, I also measure firm performance by looking at the firm’s S&P long-term domestic issuer credit rating. The rating indicates the borrower’s capacity to pay back its debt obligations and is therefore a logical measurement to test on loan sales.

Nonetheless, the shortcoming of accounting measures is their backward-looking elements and their only partial estimation of future events in terms of depreciation and amortization (see e.g., Al-Matari and Al-Swidi, 2014). To circumvent that shortcoming, I also look at market-based measurements. According to Al-Matari and Al-Swidi (2014) the advantage here is their forward-looking aspect and its consideration of investor’s future expectation. For this reason, I include Tobin’s Q as a fifth performance measurement, which proxies the market’s expectations of the firm’s ability to create value. As Tobin’s Q is defined as total market value of the firm divided by the total book value of the firm, an value greater than one indicates theoretically an overvalued firm (see e.g., Brainard and Tobin, 1968; and Tobin, 1969) However, it could also indicate growth opportunities seen by the market.

Lastly, I include the stock price of the company. This measure is widely used in previous studies (e.g., Höbarth, 2006; Lakonishok et al., 1994; Fama and French, 1992) and is easily available. It also reflects the stock market and its expectations about the future of the company. A successful future will either mean increasing future dividends or an increasing stock price, which is already priced in today.

(18)

To minimize the influence of outliers in the data, I winsorize the performance measures at the 99th percentile. However, due to larger outliers the change in stock prices is winsorized at the 95th percentile.

3.1.3.2 Lender

For measuring lender’s financial constraints, I make use of different measurements as for the borrower. The reasoning behind this is that I mainly want to focus on lenders that are financially constrained, whereas I investigate borrowers and their performance overall, but not borrowers that are financially constrained.

Beltratti and Stulz (2009) find out that larger banks with a higher tier 1 capital ratio and a higher equity ratio performed better during the financial crisis. They find a significant positive relationship between bank performance and the tier 1 capital ratio. As I argue that financially constraint and badly performing banks sell more loans, I include the tier 1 capital ratio and the equity ratio in the lender regressions as well. The tier 1 capital ratio is calculated according to the 1988 Basel Accord. The equity ratio is defined as total shareholder’s equity divided by total assets.

To also test for the performance of the lender, I use ROA (see e.g., Berrios, 2013; Demirguc-Kunt and Huizinga, 2000). ROA, defined as income before extraordinary items divided by total assets, measures bank’s ability to generate a positive net income from its investment in its assets and is therefore a proxy for profitability.

To minimize the influence of outliers in the data, lender variables are winsorized at the 99th percentile.

3.1.4 Other control variables

I control for firm and lender size measured as the book value of total assets, since larger firms might have advantages due to economies of scale and scope. This could lead to a higher market value and therefore directly influence loan sales. To control for outliers, the log of the book value of total assets is taken (see e.g., Al-Smadi et al., 2013).

Furthermore, another control variable is book leverage as a fraction of total debt to total assets. Hutchinson and Gul (2004) argue that lenders are more likely to control their investment when

(19)

the borrower is highly leveraged. More monitoring could then lead to an earlier discovery of relevant information, which influences the loan sale rate of highly leveraged loans. I also include the tangibility ratio, which is defined as tangible assets divided by total assets. As tangible assets could be pledged more easily as loan collateral, several studies show that a high tangibility ratio leads to a higher debt capacity (e.g., Shleifer and Vishny, 1992; Rajan and Zingales, 1995). In the light of this thesis, it could be that a higher tangibility also leads to a higher sales rate, as collaterals reduce the disadvantage due to information asymmetry, which might make loan selling easier.

As certain loan characteristics may influence loan sales as well, I also control for them. Therefore, the dummy variable term loan is created, indicating whether a loan tranche is a term loan or not. Drucker and Puri (2009) find out in their study that sold loans are more likely to be term loans. They reason that, in contrary to term loans, credit lines need more monitoring and require the lender to lend funds in the future and are therefore less likely to be sold (Berger and Udell, 1995). Moreover, loan characteristics include the log of the loan tranche amount, the log of loan duration until the renegotiation event measured in days and an indicator variable, covenant, taking the value of one if a covenant exists.

Moreover, certain lender control variables are added. I included an indicator variable, lead arranger, since those lenders may have superior information about the borrower and could try to exploit that. Lender commitment is the amount of debt a lender is providing within the syndicate. To distinguish between types of lenders the dummy variables bank, asset manager and CDO are added.

3.2 SUMMARY STATISTICS

Summary statistics including the most important variables used in this thesis are presented in Table II and III. Table II presents a short overview about the main dependent variables – sale rate and sale.

[Table II approximately here]

The dataset consists of 1,236 loans and 1,655 loan tranches. I was able to calculate one of the main dependent variables sale rate 1,655 times, so for every loan tranche. A sale rate of 100%

(20)

means that each lender, who was a syndicate member at origination, sold its loan share. The sale rates lies between and 0% and 100%, with an average value of 27%. However, the 75th percentile has a 50% sale rate, which may lead to the conclusion that it is a rare event when every syndicate member sells its loan share. The binary variable sale is computed for 15,743 observations. The observations are more for that variable, as I have one observation for every loan tranche – borrower – lender combination, in contrary to the sale rate variable where there is one observation for every loan tranche borrower combination. In other words, the observation are round about 9.5 times more which equals the mean amount of lenders within a syndicate (see Table III, Panel B).

Table III includes descriptive statistics at origination. To be more specific, as the loan’s origination date differs between 1996 and 2013, the results shown here are from the end of the last financial year before the origination of the loan.

[Table III approximately here]

Table III, Panel A contains an overview of borrower characteristics. The median rating equals “BB+” (=11 in my sample) on the Standard & Poor’s rating scale and is therefore one grade below investment-grade. One can observe the same result by looking at the Altman’s Z-Score. A score below 1.8 is commonly interpreted as a high probability for the company of going bankrupt, while companies with scores above 3.0 are not likely to go bankrupt. In other words, the higher the score, the lower the likelihood of bankruptcy. With a median value of 2.62 the borrowers are, generally speaking, right at the border of the safe zone to the grey zone, where the bankruptcy risk is increasing. However, 31% of the borrowers are in my defined distressed phase, indicating an Altman’s Z-Score below 1.8. I want to point out the variance in all the results, indicating an inhomogeneous sample with different borrowers.

Certain loan characteristics are shown in Table III, Panel B. Only 21% of the loan tranches are term loans. The average syndicate consists of 9.52 lenders. However, the median is 7 lenders, as some syndicates are very large with a maximum of 54 lenders. By looking at the loan amount, one can see a large difference between the smallest loan (USD 2.5 million) and the largest loan (USD 13 billion), resulting in a median loan amount of USD 245 million. This value drops to USD 100 million at the 25th percentile, indicating that the loans are granted to large- and medium

(21)

sized firms. As one can see in the summary statistics, 40% of the loan tranches have a contractually binding covenant in their loan contracts. Markup is the premium paid above a base interest-rate (mostly LIBOR). There is a large gap in the markup as well (between 0.1% and 8.5%), indicating that the sample consists of borrowers with different risk profiles. However, the mean of 1.77% is in line with an average risk profile. The typical loan duration between loan origination and the first amendment is 479 days (=16 months). Outliers drive this result, as the median here is 370 days (=12 months), which is broadly the same result as Roberts (2014) finds during his research.

Certain lender characteristics are shown in Panel C. Although there are the same amount of loans and loan tranches in this sub dataset, there are more observations. Reason for this is the fact, that there is now one observation for every loan-tranche-lender combination. 97% of all lenders are banks and only 1% are asset-managers and collaterized debt obligations. The title “lead arranger” got assigned to 36% of the lenders in the sample. It is common practice to assign more than one lead arranger for larger loans (e.g., Gopalan et al., 2011). The average tier 1 capital ratio is 9% and the average equity ratio is 7%.

I do not include summary statistics at the end of the last financial year before the amendment of the loan, as undisclosed results showed that there is no noteworthy pattern. Above that, statistics about the borrower and lender characteristics do not differ much. However, and in line with previous research, I find evidence that the characteristics of the amended loans are different to the original loans.

3.3 METHODOLOGY

This subsection discusses the methodology I use to evaluate the factors influencing loan sales and to test my hypotheses.

Hypothesis I: The performance of a borrower at origination is inversely related to sales in the loan syndicate.

To test the first hypothesis, I will use ordinary least square regressions. The dependent variable, sale rate, is a measure of loan sales in a loan tranche syndicate and therefore lies between 0% and 100%. The percentage ratio sale rate has one big advantage over the binary variable sale, namely

(22)

the fact that I only have one observation for every borrower-loan tranche combination. Therefore, I do not overweigh borrowers with numerous lenders, as I would if using the binary variable sale. To capture the effect of firm performance, I include the six performance measures described in Section 3.1.3, namely ROA, EBITDA margin, Rating, Altman’s Z-Score, Tangibility and Tobin’s Q. I control for firm size (ln (total assets)) and debt structure (book leverage) of the company. I also control for loan characteristics in further regressions.

Sale Ratei = + ₁ Borrower Performancei + Borrower Controlsi + Loan Controls + ui I expect the coefficient of my performance measures, except the rating variable, to be significant and negative. As I assigned high numbers to worse ratings, I expect the rating coefficient to be significant and positive. This result would point in the direction of my first hypothesis, as worse performing firms at origination would have a higher sale rate of their loan tranche. Furthermore, as high risky and worse performing firms can also be identified by their higher interest payments, I expect the markup coefficient to be significant and positive as well. This would mean that firms that have to pay a higher markup due to their risk profile experience a higher sale rate of their loan tranches.

Hypothesis II: The change in the borrower’s performance between origination and amendment is inversely related to sales in the loan syndicate.

To test the second hypothesis, I use the same regression method as before. The dependent variable is still sale rate, as a percentage ratio measuring the loan sales within a syndicate. However, as the first hypothesis is only looking at static variables at origination, I am now taking dynamic variables into account. Therefore, I include change variables as the main independent variables. To test for the second hypothesis, I calculate the change in performance measures between two points in time. For my baseline regression the change is calculated between the values of the last financial year before the origination of the loan tranche and the last financial year before the amendment of the very same loan tranche. However, as it is not possible with the dataset to identify the exact date of a loan sale, I alternate the two points in time for further regressions. The defined two points in time, I use for calculating the change variable, are explained in the table description above the regression results. I calculate the change in performance for the EBITDA margin, Altman’s Z-Score, rating and ROA. On top of that I also

(23)

include a measurement for the change in the borrower’s share price. Usual firm controls and loan control variables are added as well.

Sale Ratei = + ₁ Performance Change_i+ Borrower Controlsi + Loan Controls + ui The performance change for all variables, except for the rating, is computed as follows:

Performance Change_i = Performancei,t1-Performancei,t

Performance_i,t

The rating change is calculated as follows:

Performance Change (Rating)_i = Rating_t-Rating_t1

To verify my second hypothesis, I expect the performance change coefficients to be significant and negative. The change variables are defined in a way that a lower performance in t1 than in t0 leads to a negative performance change. As a result, with negative coefficients a negative performance change will support my hypothesis that the borrower’s performance is inversely related to sales in the loan syndicate.

Hypothesis III: Financially constrained lenders are more likely to sell their loan shares.

To test the third hypothesis I will use ordinary least square regressions. However, the dependent variable now is the binary variable sale. Otherwise it would not be possible to include and test for lender characteristics in the regressions. The binary variable sale takes the value of one, if a lender sold its loan share and is not in the loan syndicate at the amendment date anymore. The main independent variables (tier 1 capital ratio, equity ratio and ROA), as described in Section 3.1.3 are now measures for the performance and the financial constraints of a lender. As in the regressions for hypotheses one and two, firm and loan controls are included as well. However, because of the problem in overweighting large syndicates here, I do not focus on the magnitude of the firm control variables.

Sale_i = + ₁ Lender Performance_i+ Borrower Controls_i + Loan Controls_i+ u_i

To accept my hypothesis that financial constraint lenders are more likely to sell their loan shares, I expect the coefficients on my main independent variables to be significant and negative. This would mean that lenders with more capital (higher equity ratio and higher tier 1 capital ratio) and

(24)

more profitability (higher ROA) have a lower probability of selling their loan shares than worse doing lenders.

To capture the influence of year effects, all regressions include year-fixed effects. Undisclosed results have shown that firm- or industry-fixed effects (clustered at the same two-digit SIC industry level) do not have a significant influence on the results, thus, these are not included. Moreover, standard errors are clustered at the borrower level.

4 RESULTS

My empirical analysis starts with univariate results and continues with multivariate tests. The univariate results help to find a pattern in the data. The multivariate subsection is divided into three parts, where each part discusses results for one hypothesis.

4.1 UNIVARIATE RESULTS

Table IV shows the significance and magnitude of the differences between to sub-samples in certain borrower, lender and loan characteristics.

[Table IV approximately here]

I divide the observation into two sub samples for the borrower characteristics (Panel A) and the loan characteristics (Panel B). One sub-sample contains all observations with a sale rate below 50% (low sale rate) and the remaining loan tranches with a sale rate with or above 50% are clustered in the second sub-sample (high sale rate). To check for differences in the lender characteristics between two sub-samples, I divide the observations into two sub-samples as well. Namely one sub-sample for all observations that have a value of one for the binary variable sale (loan sold) and another sub-sample for all other not sold loan shares (loan not sold, i.e. sale=0). While looking at the borrower characteristic differences between the two samples in Panel A, one can observe that the borrower characteristics do not differ very much. However, there is a significant difference in rating, ROA and tangibility. The high sale sample has mean rating of 11.5 (Standard & Poor’s rating scale: 11 = BB+, 1 =BB) and therefore a worse rating than the

(25)

low sale sample with an average rating of 10.9. This difference is statistically significant at the 5% level.

A statistically weaker difference, at the 10% level, can be observed for the ROA: Borrowers assigned to the high sale rate sample have with an average ROA of 2.8% a lower ROA than the borrowers in the low sale rate sample with a ROA of 3.7%.

As I use a lower rating grade and lower ROA to identify worse performing lenders, those first two results of a worse rating and low ROA for high sale loans point in the direction of my first hypothesis.

A statistically strong difference on the 1% level can be found by looking at the tangibility ratio. The high sale loan tranches have with a tangibility ratio of 74.5% a statistically smaller ratio than the low sale loans with a ratio of 80.1%. One explanation could be that, as tangible assets could be pledged more easily as loan collateral, which increases the debt capacity. Furthermore, several studies show also that a high tangibility ratio leads to a higher debt capacity (e.g. Shleifer and Vishny, 1992; Rajan and Zingales, 1995) as tangible assets are more salable for creditors, even in the case of bankruptcy. The converse argument might be that loans with a lower debt capacity are structured and sold afterwards.

Although not the main focus of my thesis, the difference in loan characteristics (Panel B) is very interesting. The high sale rate sample has statistically more term loans in it (28.3% vs. 19.0%), on average two more lenders in the syndicate, a higher markup (1.94% vs. 1.71%) and a with roughly 574 days a longer duration until the amendment. The results are in line with the findings of Drucker and Puri (2009) who find out in their study that sold loans are more likely to be term loans and have a longer maturity. They reason that, in contrary to term loans, credit lines need more monitoring and require the lender to lend funds in the future and are therefore less likely to be sold (see also Berger and Udell, 1995).

Panel C shows some significant differences between the two sub-samples. In contrary to my hypothesis there is no statistical relevant difference in the lender’s ROA. As argued in the hypothesis development, the tier 1 capital ratio of the lenders in the loan sold sub-sample is marginal smaller, although not statistically relevant at the 10% level. However, the average equity ratio of 6.5% in the loan sold sub-sample is smaller as in the other sub-sample (6.7%) and statistically relevant at the 5% level. This would support the hypothesis that financially constraint

(26)

lenders sell their loan shares. The loan commitments of lenders, who sell their loan shares, are on average smaller (USD 35 million) than the loan commitments of lenders who keep their loan shares (USD 44 million). Lead arrangers also sell fewer loans, as only 17% of the lenders who sell their loan share have this title, whereas 45% of the lenders who do not sell are lead arrangers. Univariate tests illustrate that there could be a pattern that points in the direction of my hypotheses. Univariate tests, however, omit a number of determinants that could be correlated and influence each other. Therefore, multivariate analysis is used in the following to formally evaluate the magnitude of the location effect after accounting for other covariates.

4.2 MULTIVARIATE RESULTS

In this section, the results from the regressions proposed in the previous section are displayed and discussed. This section is divided into three main parts. The first part consists of the effect of certain borrower characteristics at origination on loan sales and therefore tests the first hypothesis. The second part discusses regression of the performance change variables on loan sales and is therefore used to test the second hypothesis. The section finishes with the test of the last hypothesis and the tests of effects of lender characteristics on loan sales.

4.2.1 Results for the first hypothesis

Table V presents the results of the ordinary least square regressions of the loan sale rate variable on certain borrower characteristics.

[Table V approximately here]

I run three types of regressions. Column (1) and (2) are regressions on the main variables of interest, namely the borrower characteristics from the last financial year before the origination of the original loan. To test if loan characteristics are correlated with the sale rate as well, I regress the sale rate on loan characteristics in columns (3) and (4). Columns (5) and (6) contain regressions of the sale rate on the borrower characteristics again where the loan characteristics are used as controls. The three main regressions alternately use and do not use year fixed effects. The coefficient of the borrower performance measure ROA is negative in the first two regressions, however only without year fixed effects statistically significant at the 5% level. That

(27)

implies that loan tranches of borrowers with a high ROA experience a lower sale rate. Or putting it in other words, lenders who give a loan to a borrower with a high ROA sell their loan shares less likely. The rating coefficient is statistical significant at the 5% level and positive. As described in Table I, I assigned lower ratings (rating for worse performing borrowers) a larger number (up to value 1 for S&P rating “D”). The findings here suggest a high sale rate for lower ratings. Both findings are in line with my derived hypothesis – loan tranches of borrowers with a lower performance are sold more likely.

The first two regressions also show a negative and a highly significant coefficient (at the 1% level) for the tangibility ratio, indicating that lenders who give a loan to a borrower with a high tangibility ratio sell their loan shares less likely. This finding is in line with my finding from the univariate section, where I find that high sale loan tranches have with a tangibility ratio of 74.5%, a statistically smaller ratio than the low sale loans with a ratio of 80.1% (for further explanations see Section 4.1).

The regressions on loan characteristics column (3) and (4) display, amongst other, statistically significant and positive coefficients for the variables loan duration, term loan and syndicate size. The results that term loans and loans with a longer maturity until the amendment have a higher sale rate are, as already described in Section 4.1, in line the findings of Drucker and Puri (2009). The coefficients of the markup variables are also positive and significant at the 1% level, indicating that loan tranches with a higher markup experience a higher sale rate. This finding is in line with my hypothesis, as I argued that companies, which have to pay a higher markup due to their risk profile, experience a higher sale rate of their loan tranches.

However, when regressing the loan sale rate on the borrower characteristics again while using loan characteristics as controls (columns (5) and (6)), the coefficient for the borrower performance measures ROA and rating still point in the same direction as before, but it looses significance. Nevertheless, the rating coefficient is still positive and at least significant at the 10% level and therewith supporting my hypothesis. Moreover, the coefficient of markup, also used as a performance proxy, is not statistically significant anymore.

To sum up, there are some results that point in the direction of my first hypothesis – worse performing borrowers at origination experience a higher sale rate. However, those findings loose statistical significance when additionally controlling for loan characteristics. On the one hand,

(28)

only the coefficient of the borrower’s rating (at the 10% level) and the tangibility ratio (at the 1% level) are still statistically significant. On the other hand, the coefficients of the control variables syndicate size and loan duration keep their high statistical significance. Nevertheless, there are some reasons to believe that the regression could suffer from a statistical bias, which will be discussed in Section 6.

4.2.2 Results for the second hypothesis

Table VI, Panel A and B, shows ordinary least square regressions of the loan sale rate variable on five different borrower performance change measurements, as the main independent variables and some borrower and loan characteristics variables as controls.

[Table VI approximately here]

There are two regressions for each of the five main independent variables. Those two types of regression differ in the interval used for calculating the change in performance. For the EBITDA margin, the Altman’s Z-Score, ROA and share price the change for the first regression is calculated between the last financial year before the origination of the original loan and the last financial year before the amendment event of this loan. The change interval for the second regression for these variables is between the last financial year before origination and the year of the amendment event. Those intervals are chosen, since they cover the whole lifetime of the original loan and therefore capture the complete change in performance. As described in Section 3.1.2, it is not possible to identify the exact sale date of a loan and therefore this is the best approximation. However, as the rating is published quarterly, the intervals here are for the first rating regression from the last quarter before the origination to the last quarter before the amendment event and from eight quarters before the amendment event to four quarters before amendment.

The coefficient of the EBITDA margin change (-0.0812) in column (1) is negative and statistically significant at the 1% level. This coefficient implies that loans to borrowers who experience a decrease in the EBITDA margin between the last financial year of origination and the last financial year before the amendment event have a higher sale rate. Or put the other way around, loan tranches to borrowers with an improvement in the EBITDA margin over time, have

(29)

a statistically significant lower sale rate in their syndicate. This outcome is also supported by the negative and statistically significant coefficient of the EBITDA margin change in the regressions with the alternated interval (column (2)). Both results of the main independent variables support the derived second hypothesis that the change in the borrower’s performance between origination and amendment is inversely related to sales in the loan syndicate.

However, the following regressions in columns (3) and (4) with a change in Altman’s Z-Score, as main independent variables, do not have statistically significant coefficients and therefore do not support my hypothesis.

The ROA regressions in columns (5) and (6) both have a positive and statistically significant coefficient. Those regressions must be interpreted in a way that borrowers with an improvement in ROA between the two points in time hold loan tranches with a higher sale rate in the syndicate. This outcome is truly against my second hypothesis, as the coefficient points in the other direction as expected.

Panel B of Table VI shows in columns (1) and (2) the regressions with the change in rating as main independent variables. Although there is no statistically significant result for the first regression, the coefficient in column (2) is negative (-0.0312) and statistically significant at the 5% level. My second hypothesis is supported with this outcome, as borrowers with a decrease in their rating grade experience a higher sale rate in their loan syndicated. Thus, the development of the rating grade as a measure of borrower’s performance is inversely related to the loan sale rate. Nevertheless, I do not find support for my second hypothesis by looking at the difference in the borrower’s share price between the two points in time, as those coefficients are statistically not significant at the required levels.

Though not directly linked to the second hypothesis, however remarkably, is the coefficient of the included control variables distressed, indicating as a dummy variable a Altman’s Z-Score below 1.8 in the last year before the amendment. The coefficient is negative and statistically significant (between the 1% and 10% level) in all included regressions. Apparently, loan tranches of a borrower, which is in heavy distress, have a lower sale rate than non-distressed loans. This finding is in line with the research of Drucker and Puri (2009) who find that only 3% of sold loans are in distress in their sample. However, they do not research directly on borrowers, but classify distressed loans as loans that are quoted at 80% or below par value. The coefficient

(30)

of the included control variable rating (from the last financial year before the amendment) is positive and mostly statistically significant at the 1% level in all regressions. Although just used as a control, this outcome could be interpreted in a way that the borrower’s rating at amendment is inversely related to the sale rate. This could be seen as a support of my first hypothesis, where I argued that the rating at origination is inversely related to the sale rate in the syndicate. Lastly, as already discussed in the previous section, loan tranches with a bigger syndicate and a longer duration experience a statistically significant larger sale rate.

In short, the results found here are ambiguous. On the one hand, particular when looking at the EBITDA margin change and the rating change, some results support my derived hypothesis. When measuring the change in the borrower’s performance as a change in rating and EBITDA margin, those results are indeed inversely related to sales in the loan syndicate. However, on the other hand, there are also performance measures that are either not statistically significant (Altman’s Z-Score change and share price change) or even pointing in the opposite direction of the hypothesis (ROA change). As mentioned before, there are reasons to believe that the regressions could suffer from statistical bias, which will be discussed in Section 6.

4.2.3 Results for the third hypothesis

To answer the third hypothesis the dependent variables is now switched to the binary variable sale. The dependent variable takes the value one if a lender sold its loan share and quit the syndicate, and zero otherwise. Results are presented in Table VII with the use of ordinary least square regressions of the loan sale variable on certain lender performance measures and various controls.

[Table VII approximately here]

I run three types of regressions. Columns (1) and (2) are regressions on the main variables of interest, namely the usual performance measure lender ROA and measurements of financial constraint (tier 1 capital ratio and equity ratio), and lender size (ln (lender total assets)) as a control. The values are from the last financial year before the origination of the original loan. In the following regressions in columns (3) and (4) loan characteristics are added as additional controls. Lastly, in columns (5) and (6) further borrower controls are added to the base

(31)

regressions. The three main regressions alternately use and do not use year fixed effects. To avoid multicollinearity when including the, at least on the first sight, very similar variables lender’s equity ratio and lender’s tier 1 capital ratio, I tested in undisclosed results the correlation between them. With a correlation of 0.1851 I believe it is in good order to include them in the regressions at the same time.

The coefficient of lender’s tier 1 capital ratio is negative in the first two regressions, but only significant at the 1% level with the use of year fixed-effects. However, when including the full set of controls in columns (5) and (6) this variable is still negative and beyond that highly significant at the 1% level. Consequently, even when holding constant a full set of borrower and loan characteristic controls there are statistically significant reasons to believe that the higher lender’s tier 1 capital ratio is, the less likely occurs a sale of their loan share. Or, put in other words, financially constrained lenders (i.e. lenders with a low tier 1 capital ratio) are more likely to sell their loan shares. Those findings fully support the derived hypothesis, as I argued that financially constrained lenders are more likely to sell their loan shares.

One gets the same picture when looking at the second measurement of financial constraint, the lender’s equity ratio. The coefficient of the equity ratio is even continuously negative and statistically significant at the 1% level in all six regressions. The interpretation of this coefficient is similar to the tier 1 capital ratio. As argued in the hypothesis development, the equity ratio is a measurement of the lender’s financial constraint. Thus, the results could be interpreted insofar as financial constraint lenders (i.e. lenders with a lower equity ratio) are more likely to sell their loan shares. The converse argument here is that lenders with a high equity ratio sell their loan shares less likely, insofar are the results here consistent with the third hypothesis.

The coefficients of the lender’s performance measurement ROA are somewhat ambiguous and on top of that statistically not significant on the required levels. Coherently, there is no reason to believe that lender’s with a lower profitability (lower ROA) do sell loan shares more likely, which would support the third hypothesis. However, the coefficient of the borrower’s ROA is negative and statistically significant (between the 1% and 10% level), which could be seen as a support of the first hypothesis.

When looking at the included control variables, one can see that the coefficients of the loan duration variable and syndicate size are also positive and statistically significant. As in the

On loan sales, borrower and lender conditions : evidence from the U.S.