1
The Loan Sale Decision
Which loan to sell?
University of Amsterdam Faculty of Economics and Business
MSc Business Economics, Specialisation Finance
Author: Laura van Hoek 10001985
liavanhoek@gmail.com
Supervisor: Dr. Tomislav Ladika Date: July 7, 2016
2
Abstract
The secondary market for syndicated loans is rapidly growing. The loan sales decision made by banks has been studied extensively but not from the firm performance perspective. The hypothesis that follows from existing literature is that banks want to sell the loans of badly performing firms and exploit a negative private information advantage. Banks’ liquidity management, asymmetric information and moral hazard issues in the loan sales decision are investigated in this paper. A sample of 1,582 syndicated loans issued and renegotiated between 1995 and 2003 is analysed. An average turnover of 21% per syndicate between origination and renegotiation is observed. The results suggest a negative relationship between firm performance and loan sales. The relationship is the strongest when performance is measured in the year of renegotiation. There is no evidence that banks exploit a negative private information advantage. In addition, evidence is found for a positive relationship between time to renegotiation and loan sales.
Key words: Syndicated loan, Loan renegotiation, Secondary market, Loan sale decision,
Information asymmetry, Moral hazard
This document is written by Student Laura van Hoek 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
1. Introduction
The secondary market for syndicated loans has grown rapidly over the last decades. Between 1991 and 2006, the market has increased from $8 billion to $239 billion (Drucker & Puri, 2009). Of the top 500 firms in COMPUSTAT, 90% is dependent on syndicated loan funding, financial firms excluded (Bushman & Wittenberg-Moerman, 2009). Prior research shows clear incentives for banks for why and when to sell a loan. Banks that rely on wholesale funding show more loan sales activity, the reason for this is that banks want to preserve their liquidity (Irani, Meisenzahl, & Moghadam, 2014). Banks’ need for liquidity, to meet capital requirements for example, is an obvious motivation for loan sales however information asymmetry issues can arise and limit loan sales (Drucker & Puri, 2009).
Agency problems arise between lender and borrower, because the lenders’ incentive to screen and monitor the borrower declines when secondary markets grow (Drucker & Puri, 2009). But the secondary market also creates an agency problem between buyer and seller because lenders have the incentive to sell loans expected to perform badly based on private information. Research by Berndt and Gupta (2008) shows a negative borrower’s stock performance after loan sales, where more recently Gande and Saunders (2012) find a positive borrower’s stock performance after loan sales.
The intuitive question that arises from this contradiction is whether banks sell bad loans because they have private information or not. In general, a bank can have two motives for loan sales; one is to free up capital and for risk management purposes, the other one is to exploit the negative private information advantage (Berndt & Gupta, 2008). Theoretical models can point in both directions. This paper further examines the loan sale motivational factors for banks by empirically studying the relationship between firm performance and loan sales. The research question of this paper is: Can firm performance explain banks’ loan sales decision? An additional question that I will investigate is: Do banks exploit a negative private information advantage in loan sales? Furthermore I will examine the effect of timing of performance measurement and whether there are differences between the period before and after the financial crisis.
First, I investigate the relationship between performance –in terms of accounting measures and credit rating– and loan sales. The loan sale decision has been studied extensively in prior research. Information asymmetry and moral hazard problems can explain banks’ behaviour.
4
Bushman and Wittenberg-Moerman (2009) so far have found no evidence that banks use information advantage as motivational factor for loan sales. Firm performance determines whether the borrower will be able to pay interest and repay the loan. Thus, it is reasonable to expect that a bank would include this consideration in its loan sale decision.
Second, this paper will investigate whether or not there is a difference in the relationship between firm performance and loan sales before and after the financial crisis. The relationship between reliance on wholesale funding and loan sales that Irani et al. (2014) find was not present in the period before the crisis. After the crisis, regulatory capital requirements for banks have increased heavily and it became more difficult to obtain funding. Cohen and Scatigna (2015) find evidence that for internationally operating banks capital ratios increased from 5.7% to 9.2% and leverage ratios from 2.8% to 3.7% as defined by Basel III between 2009 and 2012. They find mixed evidence for the implication of increased capital ratios for lending activity; a 1% decrease in real gross loans for banks in development countries, a 4% increase for US banks and a 7% decrease for European banks. Research by Ivashina and Scharfstein (2010) indicates that in mid-2007 the credit boom peaked and syndicated loan market activity slightly decreased but that this decrease accelerated in 2008 because banks needed to cut lending. Thus, it is plausible to expect that the loan sale decision process of the bank has changed after 2008 in order to comply with the new regulatory constraints and funding probabilities –also in relationship to firm performance– but this paper further investigates how.
The hypothesis that follows from existing literature is that banks i) want to sell the loans of badly performing firms and ii) exploit a negative private information advantage in loan sales. The answer to this question can be meaningful for policy makers/regulators and banks for example for risk management. Loans make up a large part of banks’ assets and therefore trading behaviour is very important when analysing the balance sheet. In combination with liquidity this was an important factor in recent crises.
To the best of my knowledge, this would be the first paper examining the effect of firm performance itself on the bank’s decision whether or not to sell a loan. Prior research focusses on bank characteristics and macroeconomic factors as determinants of loan sales. The ex-post performance, in terms of stock price development, is studied by Berndt and Gupta (2008), Gande and Saunders (2012) and Dahiya, Puri and Saunders (2003) amongst others, not the ex-ante motivation. This paper will answer which loans are sold by banks,
5
where existing literature answers why and when loans are sold. The second contribution of this paper is that it provides insight into the effects of the financial crisis on the secondary market for syndicated loans.
This paper is based on a data set of 1,582 renegotiated syndicated loan facilities, issued by banks between 1995 and 2013 to US publicly listed firms. This data from Reuters’ Thomson ONE is combined with accounting data from the COMPUSTAT S&P Historical Annual Fundamentals database and S&P credit ratings, also available via COMPUSTAT. I will perform empirical analysis using several regression methods. First, I define a loan sale variable and multiple performance variables. OLS regressions are applied using these turnover and performance variables and later control variables for loan characteristics and borrowers’ characteristics are added. After that, fixed effects regression and random effects regression models are estimated. Finally, this paper estimates the model for a subsample using only loans that are issued after the beginning of the financial crisis, from 2008-2013. The first results report that there is an average turnover of 21% per syndicate. Also, the number of banks in the syndicate increases between origination and renegotiation from 18 to 23. The average time to renegotiation for the amended loans is 20 months. I have found some evidence that firm performance is negatively correlated with loan sales activity. Regression analysis shows a significant negative coefficient for EBITDA/Total Assets and Interest Coverage ratio in the different estimated models. Thus, firms with a lower reported EBITDA/Total Assets or Interest Coverage Ratio show a higher turnover on their loans. The impact of performance is measured at four moments. The first moment is defined by the results reported in the issue year, the second measure is the reporting date one year later, both numbers are a proxy for performance at the time the loan is provided to the borrower. The third measure of performance is in the year of renegotiation and the fourth one year after renegotiation. Economically, a lower EBITDA/Total Assets of 10% measured at the issue year is associated with a 1.8 percentage points increased turnover. Interestingly, this negative relationship is the strongest for performance measured at the year of renegotiation. A lower EBITDA/Total Assets of 10% measured at the year of renegotiation increases turnover with 3.1 percentage points.
Consistent with this finding, firms with lower Interest Coverage Ratio also show a higher turnover on their loans. However, the economic impact of the Interest Coverage Ratio is very limited. The coefficients for credit rating, which reflect the chance of default, are significant
6
as well, which means that credit rating is a determining factor in the bank’s loan sale decision. With regards to the industry of the borrowing firm, being active in the financial industry decreases turnover with 16 percentage points. In addition, I have found that duration is positively correlated with loan sales. In this paper, duration is measured as the time between origination and renegotiation. Thus, the longer it takes a firm to renegotiate the loan contract, the higher the turnover. The results affirm the first part of the stated hypothesis. The findings do not document a relationship between firm performance measured at the year after renegotiation and loan sales. This suggests that there is no evidence that banks exploit a private negative information advantage in loan sales. Hereby, the results do not affirm the second part of the stated hypothesis.
Subsample results for after-crisis issued loans differ from the full sample results in economic magnitude and timing of performance measurement. The effect of EBITDA/Total Assets on turnover is stronger for the period after 2008 than for the full sample. A 10% lower EBITDA/Total Assets measured in the year of issuance increases turnover with 5.0 percentage points, relative to 1.1 percentage points for the period before 2008. After 2008 performance measured at the issue year is of greater impact than performance measured at later stages. Again, there is no evidence for a relationship between performance after renegotiation and loan sales.
The remainder of this paper is structured as follows. In the second part a background on syndicated loans is provided and related literature is reviewed, the third part describes the data and its sources. In the fourth part, the method, results and robustness checks are presented. In part six, the conclusion and ideas for future research follow.
7
2. Literature review
2.1 Background information
A syndicated loan is a joint loan provided by more than one financial institution to a firm (Sufi, 2005). Syndicated loans are private debt instruments but have commonalities with public debt because they have a credit rating and are tradeable on a secondary market (Bushman & Wittenberg-Moerman, 2009). The market for syndicated loans can be divided into two categories; the primary syndicated loan market and the secondary loan sales market (Dahiya, Puri, & Saunders, 2001).
2.1.1 Primary syndicated loans market
There are two types of lenders within each syndicate: lead arrangers and participant lenders (Sufi, 2005). The loan is provided under one common contract with equal terms for all members of the syndicate but lead arrangers usually hold a larger share than participant arrangers. Lead arrangers are responsible for the direct contact and relationship with the borrower. They are also primarily responsible for collection of information, governing terms, enforcing of restrictive covenants, calculation of interest and administration of drawdown of funding of the borrowing firms (Sufi, 2005). Renegotiation of the loan can only be established with unanimity of all lenders in the syndicate, both lead arrangers and participant arrangers (Sufi, 2005).
The first step in the syndication process is for the lead arranger and the borrower to sign a preliminary loan agreement (Sufi, 2005). This is the so-called ‘mandate’. In the preliminary loan agreement the sum of the loan, a range for the interest rate, restricting covenants, collateral and applicable fees are described (Sufi, 2005). After that, the lead arranger approaches potential participants with an Information Memorandum on the borrower. The selection process of participating arrangers is often complex and untransparent but usually the lead arranger makes a selection of candidates and these potential participants can choose to accept or reject the offer, this process will be further described in the next paragraph. After the participant lenders have accepted and reached an agreement with the lead arranger on the part of the sum that is funded, all lenders sign the final loan contract. The contract terms are equal for all parties, both the lead arranger and participating arrangers (Sufi, 2005). Finally, a fee for managing the syndicate is paid by the borrower to the lead arranger next to the commitment fee (Sufi, 2005).
8
The composition of the group formed to provide the loan is the outcome of multiple actions by the lead arranger, participating arrangers and the borrower (Lee & Mullineaux, 2004). The main variable is the size of the syndicate. The syndicate will be larger when lenders prefer to participate in small fractions of a loan. Smaller syndicates, on the other hand, enable a more efficient managing problem, which makes it easier to act promptly in times of financial distress.
The lead arranger cannot solely control the size of the lending group. The composition of the lending group affects the incentives to screen and monitor the borrower (Lee & Mullineaux, 2004). There is very limited monitoring in the bond market after the issue date, partially because bonds are similarly to syndicated loans held by multiple members, but for the syndicate market this may be different (Lee & Mullineaux, 2004). In existing literature it is proven that composition responds to information asymmetry costs and default probability, thus, it can be concluded that screening and monitoring is important in the syndicated loan market (Lee & Mullineaux, 2004).
Lee and Mullineaux (2004) find that smaller and more concentrated syndicates are formed when little information is available about the borrower, credit risk is relatively high and when loans are secured. On the other hand, large and more diffuse syndicates are formed when maturity of loans is relatively longer, in line with existing literature evidence that long term loans have less credit risk, and when reputable lead arranging banks are involved. In addition, they find evidence that when loan sales is restricted, larger syndicates are formed as well.
2.1.2 Secondary syndicated loans market
Sufi (2005) explains the key differentiations between the primary and secondary market for syndicated loans. The most important difference is that a loan sale, secondary market, leaves the contract terms between lender and borrower the same, while these can be changed in a new loan agreement, primary market. The secondary loans market has grown more rapidly than the primary syndicated market. From 1991 to 2006, with a compounded annual growth rate of 25%, secondary market volume increased from $8 billion to $239 billion (Drucker & Puri, 2009). Drucker and Puri (2009) also find that 60% of all syndicated loans is sold within one month from the issue date and 96% within a year.
In the secondary syndicated loans market, or loan sales market, members of the syndicate can sell their loan over the counter via assignment or participation, most of the time through
9
dealers at large underwriting banks (Bushman & Wittenberg-Moerman, 2009). After assignment sales, the buyer is a direct syndicate member. In sales via participation, the buyer only receives a participating interest in the original holder’s commitment (Bushman & Wittenberg-Moerman, 2009). The majority of sales in the syndicated loans market is done via assignment.
2.2 Related literature 2.2.1 Loan sales motivation
As noted in the introduction, the objective of this paper is to examine whether there is a relationship between firm performance and loan sales. In order to answer this question, it is important to investigate the potential motivational factors for a bank to sell the loan. Banks’ need for liquidity is an obvious motivation for loan sales, however information asymmetry issues can arise and limit loan sales (Drucker & Puri, 2009). Agency problems arise between borrower and lender, because the lenders’ incentive to monitor and screen the borrower declines when secondary markets grow (Drucker & Puri, 2009). Loan contracts can be arranged in a way to solve for this problem, however, if no incentive to do so and great market liquidity is present, little effort will be put into this. The secondary market also creates an agency problem between buyer and seller because lenders have the incentive to sell loans expected to perform badly based on private information (Drucker & Puri, 2009).
Irani et al. (2014) find evidence that bank liquidity risk management is indeed a motivational factor for loan sales. They use a US credit register that is administered by regulators to examine secondary market activity of loan syndicates. They prove that there is more loan sales activity amongst banks that rely on wholesale funding and argue that the reason is that those banks want to preserve liquidity. This effect however was not observed for the period 2003 until 2006, the reason for this could be that pre-crisis it was easier for banks to obtain funding. Over the last few years bank have become more reliant on wholesale funding, which makes banks more sensitive to funding shocks. Irani et al. (2014) conclude that banks reliant on wholesale funding use loans sales as a method to smooth out these funding shocks. This is not supported by earlier literature. Bank funding and its relation to financial stability has proved to be of significant impact during the last crisis, so further empirical research is needed for complete understanding.
10
2.2.2 Loan renegotiation
Sufi (2005) investigates renegotiation considerations of banks and the resulting structure of the syndicate. Renegotiation can include a change in the terms of the principal, interest, time to maturity or collateral for the loan amongst other characteristics.
Roberts and Sufi (2009) find that 75% of all loans are renegotiated on maturity, principal or interest before its maturity date. For contracts with maturities of over one year or three year, this increases to 90% and 96% respectively. Roberts and Sufi (2009) examine what information triggers banks to renegotiate. The outcome presents many variables that impact renegotiation including: credit quality information, decreasing costs of equity, macro factors, credit market conditions, lender characteristics and stock markets. Initial loan contract terms seem to have little impact on renegotiation.
Bolton and Scharfstein (1996) conclude that the optimal size of the syndicate balances two opposite effects. A borrower that is more likely to strategically default should experience more difficulty in renegotiation because the bank wants to discourage poorly performing management. On the other hand, when a borrower is more likely to default because it is sensitive to exogenous shocks, a smaller group of lenders makes renegotiation easier. Following these two predictions, Sufi (2005) tests the following question: is liquidity default or strategic default the more important factor in syndicate renegotiation?
The research starts with the assumption with the assumption that unanimity of lenders in the syndicate is required to renegotiate on the terms of the loans (Sufi, 2005). So called “participant-loading” contributes to a decrease of the expected payoff from renegotiation for the borrower. Hereby, an increase in the number of lenders in the syndicate leads to a decrease in the chance of strategic default by borrowers. In line with the incomplete contract theory, Sufi (2005, p.5) states “the ex-post renegotiation considerations affect ex-ante contract structure”. The conclusion is that the effect of strategic default is stronger which indicates that lenders indeed add participants to the syndicate to make renegotiation harder when the borrower is in default (Sufi, 2005).
2.2.3 Originate-to-distribute model
The rapid growth of the secondary market for syndicated loans has also raised concerns about the bank’s special role of monitoring (Bushman & Wittenberg-Moerman, 2009). Bank lending has shifted from the traditional model to this originate-to-distribute model where
11
banks can issue loans and sell the risk to investors who are willing to take it. They still earn their origination fees but are less exposed to credit risk. This has led to an active debate involving banks, borrowers, investors and regulators on the so called originate-to-distribute model in lending fuelled by the economic crisis in 2007 (Bushman & Wittenberg-Moerman, 2009). Banks have superior information on the borrower which could lead to information asymmetry issues. As an example of the classic Lemons’ Theorem this would lead to only bad loans being sold in the secondary market. Whether regulators should limit this originate-to-distribute model depends on the level of value creation and value destruction that results from the policy (Berndt & Gupta, 2009).
There are examples where loan sales can be completely reasonable for a bank. When loans are sold for capital relief, the bank could turn to more profitable activities or use it for more loan origination for which they can earn a fee (Berndt & Gupta, 2009). As a consequence, return on equity and return on assets will increase. This is a comparative advantage of banks over other financial institutions. Besides, loan sales can have diversification purposes and third, increased liquidity leads to lower costs of capital for the banks (Berndt & Gupta, 2009). These three arguments justify the originate-to-distribute model, as opposed to the information asymmetry and moral hazard issues.
From the borrowers’ point of view, a loan sale can have a positive impact on the firm by lowering its costs of capital, increasing its access to funding and increasing the availability of information on the market. There are also possible negative implications for the borrower resulting from a loan sale. Amongst the negative effects are higher covenants for more traded firms, deterioration of the relationship with the originating bank and unintended risk-shifting by management as a result of decreased monitoring (Berndt & Gupta, 2009).
2.3 Other loan sales related literature 2.3.1 Stock market reaction
As highlighted in the introduction, Gande & Saunders (2003) find a significant positive reaction of stock price following loan sales. Both negative and positive effects of loan sales on equity holders of the borrower can be identified and therefore empirical research results point in both directions.
A potential benefit of a loan sale as perceived by investors is that the event can ease financial constraints of the borrowing firm (Gande & Saunders, 2003). Hereby wealth is transferred
12
from bondholders to equity holders. By increased liquidity or because of risk-sharing benefits the cost of capital of the borrower can also decrease. A potential negative effect that this could lead to risk-shifting by management and indirectly negatively affect equity holders. Gande & Saunders (2003) also find a positive reaction of stock price after a new loan announcement. Berndt & Gupta (2009) find the contradicting result that stock returns of firms with actively traded loans significantly underperform those that are not actively traded. For a period of three years after the trading, they find a underperformance of 8-14% for firms of actively traded loans.
2.3.2 Information asymmetry and moral hazard
Prior research by Bushman & Wittenberg-Moerman (2009) examines whether the information asymmetry is exploited by banks, which would lead to the sale of low quality loans only on the secondary market, or not. The results indicate that there is no discount related to loan sales as exploitation of information advantage. After the initial sale, loan prices decline rapidly but differently for the subsamples. Differences can be attributed to reputation of the arranger and purpose of the loan (e.g. LBO/MBO).
Moral hazard regarding loan sales is related to the incentives to screen and monitor. The incentive to screen the borrower is greater when a bank holds a loan relative to when it sells the loan (Duffee & Zhou, 2001). For potential buyers of a loan there is no information about the extent of monitoring by the originating bank, which creates room for moral hazard (Gorton & Penacchi, 1995). Duffee & Zhou (2001) describe the mechanism where on the one hand banks preserve their own control over monitoring of borrowers by limiting loan sales activity and on the other hand the value of monitoring that can be offset by the value of loan sales. In equilibrium, commitment to monitoring is desirable for the bank taken deadweight costs into account.
13
3. Data
3.1 Sample selection
A large data set of syndicated loans provided to US publicly listed companies is retrieved from Reuters’ Thomson ONE1. The data set contains information about the loans on loan share level including the variables borrower, lender, issue date, maturity date, facility identifier and tranche identifier for the period 1995-2013. A second data set links original loans and corresponding amended loans to each other. By merging this information, the changes in ownership in the syndicate between origination and the amendment date can be identified.
Lenders in the data set are selected by their type of business. Financial institutions other than banks are excluded: asset management, hedge funds, insurance companies, life insurers a.o. were manually identified based on judgement of their names. 3.153 renegotiated loan shares were deleted because they were issued by non-bank financial institutions and 142 loan shares were deleted because the lender type could not be identified based on the information available in the data set.
Borrower information is collected from COMPUSTAT. All annual fundamentals are available in the S&P Historical Annual database. This information is matched to the syndications data set by a company identifier (gvkey) and year.
The dependent variable loan sale is defined as the percentage of lenders that changed ownership between issue and amendment per tranche of the facility. There is no information about the changes in the composition of the syndicate in between, but this data set possibly already can explain more than existing research.
3.2 Variables
1. Turnover variable
The turnover variable is used to express the dependent variable loan sales. Turnover is defined as the percentage of the syndicate that has changed ownership between issue and renegotiation: number of banks that changed ownership/number of banks in the syndicate at
14
origination *100. In the data set a syndicate is jointly defined by a facility identifier and a tranche identifier.
2. Performance variables
For firm performance, several variables are defined. First, in line with research by Bushman and Wittenberg-Moerman (2009), accounting performance measured as EBITDA/Total Assets is calculated. What matters most for banks is whether the borrower is able to pay interest and repay the principal. Performance measures that are relevant for interest payments and repayment are Interest Coverage Ratio, defined as EBIT/Total Interest Expense, and chance of default, which can be measured by credit rating. The Domestic Long Term Issuer rating is defined by Standard & Poor’s as “The likelihood of payment - the capacity and willingness of the obligor to meet its financial commitment on a financial obligation in accordance with the terms of the obligation (…)” (Standard & Poor’s).
For relevancy and availability reasons and comparability with existing literature, EBITDA/Total Assets, Liquidity Coverage Ratio and credit rating are the performance variables that are used in estimation of the model. Under the hypothesis that banks want to sell the loans of badly performing firms, the correlation with turnover is expected to be negative.
The performance variables EBITDA/Total Assets and Interest Coverage Ratio are defined at four moments. The first measurement is indicated by the reported results in the year of issuance. The second moment is measured the year after the issue year. The third measurement is indicated by the reported results in the year of renegotiation and the fourth in the year after. If banks exploit their negative private information advantage, firms with worse performance after renegotiation are expected to have higher turnover.
3. Loan variables
Loan variables are proxies for qualitative and quantitative characteristics of the loans. Two variables are included in estimation of the model: the number of banks in the syndicate and the duration of the loan. Flannery (1986) found that mainly low-quality firms issue long term debt. However, the correlation coefficient matrix shows no evidence for problematic correlation between the performance and duration variables.
15 4. Borrower characteristics
Besides firm performance, other borrowers’ characteristics are used as independent variables in the model. The logarithm of Total Assets is used to control for the effect of firm size. Lager firm size could lead to more loans outstanding and larger liquidity and therefore higher turnover. Leverage is measured as Long Term Debt/Total Assets and is also included in the model. Finally, the industry of the borrowing firm as defined by the Global Industry Classification Standard (GIC) industry sector codes is included in the model.
Correlation between the variables specified is reported in Table IV.
Table IV Correlation Coefficient Matrix
EBITDA/ Total Assets Interest Coverage Ratio LT Debt/ Total Assets Total
Assets #banks Duration EBITDA/ Total Assets 1.0000 Interest Coverage Ratio 0.2759 1.0000 LT Debt/ Total Assets -0.1877 -0.2147 1.0000 Total Assets -0.0711 -0.0407 0.0290 1.0000 #banks -0.0176 -0.0409 0.1617 0.3325 1.0000 Duration 0.0527 -0.0003 0.0467 -0.0111 0.1302 1.0000 3.3 Descriptive statistics
Table I on page 17 summarizes the key statistics of the variables. The first observation from the descriptive statistics for loan characteristics is that the number of banks in the syndicate differs for Original issued loans and the Amended loans. The mean increases from 18 banks per facility at origination to 23 banks per facility at amendment. A two-sample t-test shows that the mean difference of Original and Amended is significantly different from zero at a 1% level. According to Lee and Mullineaux (2004), syndicates are usually larger when credit risk is relatively low or a reputable lead arranger is involved. Duration increases due to the fact that Original loans are renegotiated before their stated maturity, while duration of the amended loans reports the number of months between renegotiation and newly stated
16
maturity. The average duration of renegotiated loans is 42 months, which corresponds to a little less than four years. The t-test shows a significant difference but this is due to the fact that the definitions for Original and Amended loans differ. The t-test for commitment does not show a significant difference, but this can be attributed to the lack of data on commitment.
Table II on page 17 reports statistic details of the loan sale variable for all amended loans. The average change in ownership per tranche per facility is 21%. The distribution of turnover is graphed in Figure I and the change in turnover over time in Figure II, to be found in the Appendix. The average time to renegotiation, as measured by duration, is 20 months. The change of duration over time is graphed in Figure III in the Appendix. This means that on average per tranche 21% of the loan shares is sold, at least once, within 20 months. This is not in line with Drucker and Puri’s findings (2009), they found that 96% of all loans are sold within one year, which might be affected by the fact that this is a sample of amended loans. In the discussion section, the possible problems will be discussed.
Note that firms lacking sufficient financial and loan level information are excluded. A mean EBITDA/Total Assets of 13% is reported and a leverage ratio for Long Term Debt/Total Assets of 29%. The main performance indicators for the remainder of this paper are EBITDA/Total Assets and Interest Coverage Ratio. Advantages of these ratios over other performance measures are that it is publicly available after financial reporting, in the short term not affected by the loan sale decision and it changes over time. The mean number of banks per facility is 17 for the used sample. The average leverage ratio for firms in the sample is 29%, measured as Long Term Debt/Total Assets.
Table III on page 18 reports the credit rating statistics for all amended loans. Credit rating scores used are S&P Domestic Long Term Issuer ratings retrieved from the COMPUSTAT ratings database. Defaults are retrieved from the S&P Historical Annual database in COMPUSTAT as Research Co Reason for Deletion (2=Bankruptcy). Credit rating and the mean ownership change per category are presented. The first observation from this table is that the highest turnover belongs to companies with credit rating CCC and the lowest turnover belongs to companies with credit rating AA-. De majority of loans in this sample is rated in the categories between BBB and B+.
Table IV on page 18 reports statistics on the GIC industry sector codes (MSCI). This classification categorizes all publicly listed companies into 10 sectors and is maintained by
17
MSCI and S&P. Data is available in the S&P Historical Annual database in COMPUSTAT too. The majority of the amended loans is provided to firms active in the sectors Consumer Discretionary, Industrials and Energy. The highest turnover of 24% is attributed to Consumer Discretionary firms and the lowest turnover of 9% to Financials.
Table I Summary Statistics
Panel A Descriptives
Descriptive statistics for all loans, at origination and amendment for the period 199-2013 Distribution
N Mean SD P25 Median P75 Min Max
Original loans
#banks per facility 1585 17,93 20,57 6,00 12,00 22,00 1,00 225,00 Duration 1516 19,47 15,58 9,50 14,00 24,00 0,00 98,00 Commitment per bank 8284 37,61 56,29 11,67 23,70 45,00 -60,00 950,00
Amended loans
#banks per facility 1582 22,85 27,10 9,00 15,00 26,00 1,00 359,00 Duration 1576 41,92 20,89 23,50 46,00 60,00 1,00 119,00 Commitment per bank 8760 38,47 68,81 11,25 23,95 40,00 -8,72 913,46
Panel B T-tests
T-test for differences between Original and Amended loans N Df t P(|T|>|t|) #banks per facility 3167 2.926 55.351 0.000
Duration 3092 3.090 -340.428 0.000
Commitment per bank 17044 16.969 -14.061 0.160
Table II Summary Statistics
Descriptive statistics for all amended syndicated loans for the period 1995-2013
Distribution
N Mean SD P25 Median P75 Min Max
Turnover variable
% ownership change 1146 20.54 22.00 0.00 15.38 30.43 0.00 100.00
Performance variable
EBITDA/Total Assets 1141 0.13 0.08 0.09 0.12 0.17 -0.44 0.53 Interest Coverage Ratio 1118 10.45 45.56 1.89 4.86 8.81 -140.22 1161.62
Loan variables
#banks per facility 1146 16.82 18.82 6.00 11.00 21.00 1.00 225.00 Duration in months 1095 20.04 15.64 10.00 14.00 25.00 0.00 98.00
Borrower variables
Total Assets 1145 7.43 1.54 6.28 7.35 8.40 3.23 12.30 Long Term Debt/Total Assets 1144 0.29 0.17 0.16 0.27 0.39 0.00 1.60
18
Table III Credit rating descriptives
Turnover numbers report mean per groupCredit Rating Observations Defaults Turnover(%)
AAA 10 0 15.66 AA+ 5 0 21.11 AA 16 0 20.66 AA- 22 0 15.23 A+ 34 0 16.71 A 60 2 17.63 A- 52 1 15.53 BBB+ 39 0 19.74 BBB 73 0 19.97 BBB- 72 3 23.87 BB+ 29 0 17.76 BB 98 2 23.51 BB- 161 0 22.08 B+ 173 4 23.17 B 64 0 20.14 B- 10 0 17.17 CCC+ 6 0 26.53 CCC 1 0 6.25 CCC- 1 0 25.00
n/a 221 n/a n/a
Table IV GIC industry sector code descriptives
Turnover numbers indicate mean per groupGIC Sector Observations Turnover(%)
10 Energy 186 20.82 15 Materials 109 20.00 20 Industrials 222 19.60 25 Consumer Disc. 290 24.04 30 Consumer Staples 68 17.68 35 Health Care 90 19.74 40 Financials 14 8.58 45 Information Technology 76 19.71 50 Telecommunication Services 16 22.91 55 Utilities 63 15.51
19
4. Method and results
4.1 Regression analysis
In the first part of the regression analysis estimates of the full sample are presented; OLS regressions, entity fixed effects, time fixed effects and random effects models are estimated. In the second part, subsample estimates are presented for the loans issued before and after the start of the financial crises.
4.1.1 Full sample estimates
The main variables of interest are EBITDA/Total Assets and Interest Coverage Ratio. Control variables are added to control for loan- and firms characteristics. First, various OLS regressions are performed with respect to the loan sale variable using performance measured in the year of issuance. All results of the OLS regressions are summarized in Table V on page 22. In columns (1), (2) and (3) the coefficient for EBITDA/Total Assets is statistically significant at a 5% or 10% level. The performance coefficient for EBIDA/Total Assets also has a negative sign, in line with the hypothesis. However, the presence of omitted variable bias could be problematic for the strength of the interpretation of these results. Potential presence of omitted variable bias is further addressed in section 4.2. This analysis suggests a negative relationship between firm performance and loan sales. There is also some evidence for a significant relationship between duration of the loan, in this case the time between origination and renegotiation, and loan sales.
In column (4) dummy indicator variables for credit rating are included in the regression. Ratings are divided into three groups, A, B and C, because the separate ratings have too few observations. AAA to A- fall in category A, BBB+ to B- fall in category B and CCC+ to CCC- form credit rating category C. From AAA to CCC- the credit ratings express creditworthiness in descending order. With a P-value of 0.052 this credit rating variable is of significant influence at a 10% level on the loan sale decision, again in line with the hypothesis. Now, the coefficient for Interest Coverage Ratio is, however small, also significant and negative in line with the hypothesis. In column (5) dummy indicator variables for the borrower’s industry are included in the regression. With a P-value of 0.502, the combined effect of difference in industry is not significant for the loan sale decision. The only industry variable coefficient that is significant is the dummy for Financials, borrowers in this industry are associated with a 16% lower turnover.
20
The results also provide evidence for a positive relationship between duration and turnover. The duration variable for these amended loans is defined as the time in months between origination and renegotiation. From column (3), (4) and (5) there can be concluded that the longer the time between issue and renegotiation, the higher the turnover. This is in line with previous research and the hypothesis. Banks make renegotiation harder when performing badly, therefore longer duration could be an indirect indicator of bad performance too.
The equation of the OLS model as estimated in column (5) is expressed below. 𝐿𝑜𝑎𝑛 𝑆𝑎𝑙𝑒𝑖,𝑗,𝑡 = 𝛼𝑖,𝑡+ 𝛽 𝐸𝐵𝐼𝑇𝐷𝐴 (𝑇𝑜𝑡𝑎𝑙 𝐴𝑠𝑠𝑒𝑡𝑠)𝑖,𝑡+ 𝛽 𝐸𝐵𝐼𝑇 (𝐼𝑛𝑡𝑒𝑟𝑒𝑠𝑡 𝐸𝑥𝑝𝑒𝑛𝑠𝑒)𝑖,𝑡+ 𝛽( 𝐿𝑜𝑛𝑔 𝑇𝑒𝑟𝑚 𝐷𝑒𝑏𝑡 𝑇𝑜𝑡𝑎𝑙 𝐴𝑠𝑠𝑒𝑡𝑠 ) + 𝛽 ln(𝑇𝑜𝑡𝑎𝑙 𝐴𝑠𝑠𝑒𝑡𝑠) + 𝛽𝐷𝑢𝑟𝑎𝑡𝑖𝑜𝑛 + 𝛽#𝑏𝑎𝑛𝑘𝑠 + 𝛾𝑅𝑎𝑡𝑖𝑛𝑔𝐵+ 𝛾𝑅𝑎𝑡𝑖𝑛𝑔𝐶+ 𝛾𝐺𝐼𝐶15+ (… ) + 𝛾𝐺𝐼𝐶55+ 𝜀𝑖,𝑡
For column (6) and (7) the regressions are performed using firm performance measured respectively in the year after issuance and the year of renegotiation. My objective is to examine whether the timing of performance measurement matters. Now, the magnitude of the coefficient for EBITDA/Total Assets has almost doubled and the significance increased from 5% to 1%. The turnover variable is expressed as a percentage, e.g. 20 indicates 20% turnover within the syndicate, where EBITDA/Total Assets is expressed as a fraction. Economically this means that a 10% lower EBITDA/Total Assets in the issue year increases the turnover until renegotiation with 1.8 percentage points, where the same lower EBITDA/Total Assets in the renegotiation year increases the turnover with 3.1 percentage points. This result indicates that loan sales is stronger related to performance in the year of renegotiation -renegotiation on average takes place 20 months after issuance- than performance in the issue year. Unfortunately, based on this data it is unknown when the loans are sold, so the banks could either have sold just before renegotiation based on actual performance or in an earlier stage based on private information about performance in the future.
With regards to Interest Coverage ratio, firms with a 10% lower Interest Coverage Ratio in the renegotiation year experience a 0.1 percentage points higher turnover. Although significant, the economic impact of this ratio on loan sales is very limited. Based on the fact that this ratio explains the ability of the borrower to pay interest I expected a greater effect. A possible explanation could be that interest payments are not the main concern of the bank, but
21
the repayment of the principal is. Repayment of the principal is more closely linked to performance in general and especially the chance of default.
After that, firm performance measured in the year after renegotiation is included in the regression to test whether banks exploit a negative private information advantage. Column (8) documents that the coefficients for both EBITDA/Total Assets and Interest Coverage Ratio are no longer significant. Thus, since there is no evidence that worse performance after renegotiation is associated with a higher turnover I cannot conclude that banks exploit a negative private information advantage. This confirms the findings of Bushman and Wittenberg-Moerman (2009), who found no evidence for higher performance of traded loans versus non-traded loans in a period of three years after the loan sales.
Now, the second explanation for when the loans are sold by the bank is less plausible since there is no evidence for a relationship between performance after renegotiation and loan sales. Based on this assumption, the results do not provide evidence for the shift to an originate-to-distribute model by banks (Berndt & Gupta, 2009). Banks do not issue loans with the predetermination to sell it, but when at some point in time for other motivational factors such as need for liquidity loan sales is desired, they decide to sell the loans of badly performing borrowers.
The results are not in line with findings of Drucker and Puri (2009), that suggest that 60% of syndicated loans is sold within one month from origination. In addition, they suggest that syndicated loans with tighter covenants are associated with higher turnover. Therefore, an explanation for this difference could be that the loans in this sample have covenants less tight than their sample. Unfortunately, data on covenants of these amended loans is not available.
22
Table V Regression Analysis
Independent variable: turnover, performance variables measured at different times, (0) indicates issue year, (1) indicates year after issuance, (R0) indicates renegotiation year, (R1) indicates year after renegotiation
OLS
(1) (2) (3) (4) (5) (6) (7) (8)
EBITDA/Total Assets -17.51** -18.01** -19.36** -18.36** -18.05* -27.25*** -31.47*** -10.36
(8.003) (8.081) (8.115) (9.226) (9.384) (7.044) (7.081) (8.769)
Interest Coverage Ratio -0.018 -0.018 -0.015 0.026 0.024 0.098*** 0.123*** -0.012
(0.015) (0.015) (0.014) (0.042) (0.042) (0.034) (0.036) (0.077)
Long Term Debt/Total Assets 0.974 0.556 -0.200 0.953 3382 5225 3.642
(3.92) (3.94) (4.43) (4.59) (4.49) (4.52) (4.83) Total assets -0.711* -0.304 -0.026 0.161 -0.026 -0.014 0.273 (0.428) (0.455) (0.564) (0.595) (0.594) (0.595) (0.620) Duration 0.289*** 0.249*** 0.245*** 0.265*** 0.272*** 0.279*** (0.043) (0.045) (0.045) (0.046) (0.048) (0.050) #banks -0.030 -0.040 -0.037 -0.028 -0.032 -0.045 (0.036) (0.036) (0.037) (0.037) (0.037) (0.038) Constant 23.01*** 28.08*** 19.96*** 14.70*** 12.63** 13.43** 13.15** 9.773 (1.206) (3.688) (3.908) (5.587) (6.118) (5.957) (5.967) (6.276)
Categorical variable "Rating" No No No Yes Yes Yes Yes Yes
Categorical variable "GIC" No No No No Yes Yes Yes Yes
Performance measure at T= 0 0 0 0 0 1 R0 R1
N 1107 1106 1056 867 865 865 855 8.21
Adjusted R-squared 0.006 0.007 0.044 0.044 0.045 0.063 0.069 0.064
Standard errors in parentheses * p<0.10, ** p<0.05, *** p<0.01
23
To control for entity fixed differences between loans, fixed effects regression is applied. Note that the data set is a weakly balanced panel data set because not all borrowers have loans issued in all years. The facility identifier is labelled as the group variable and the issue year is labelled as the time variable. After that, year fixed effects are also included. Binary dummy variables for each year are added to the equation, which results in T-1 time periods. Whether fixed effects regression is the appropriate analysis for this data set will be investigated in a later stage.
Estimates of the fixed effects models are presented in Table VI on page 24. Columns (1), (2) and (3) indicate significant coefficients for EBITDA/Total Assets and no significant coefficients for Interest Coverage Ratio. Columns (2) and (3) show that there are significant coefficients for loan variables and borrowers’ characteristics. Again, negative coefficients for the EBITDA/Total Assets are reported. Both significance and magnitude are smaller for the performance variables relative to the OLS estimates.
A testparm test is performed to verify whether time fixed effects are needed in the fixed effects model. The P-value shows that at a 10% level the null hypothesis that the coefficients for all years jointly are equal to zero is rejected. A Hausman test is performed to decide between fixed effects or random effects regression analysis. This test indicates that the unique errors are not correlated with the regressors and therefore random effects should be the preferred model at a 5% level. Table VII on page 24 reports the estimates for the random effects regression.
Not all fixed effects and random effects assumptions hold. 54 firms have signed multiple loans on the same date and therefore have to drop out in this analysis. Besides that, the number of firms signing loans in multiple years is also limited in the sample, therefore the number of observations and the strength of the results decrease.
To summarize the full sample results, the OLS estimates provide evidence that banks want to sell the loans of badly performing firms, especially for performance measured in the year of renegotiation, as stated in the first part of my hypothesis. The results provide no evidence that performance after renegotiation is related to loan sales and therefore the second part of my hypothesis cannot be affirmed. FE results provide the same evidence, however these findings should be interpreted carefully because not all assumptions hold.
24
Table VI Fixed Effects Regression
Independent variable: Turnover; performance measured at renegotiation year (R0)
Fixed effects
(1) (2) (3)
EBITDA/Total Assets -34.20** -36.05** -29.78* (14.91) (14.91) (15.62) Interest Coverage Ratio -0.024 -0.013 -0.060 (0.131) (0.130) (0.135)
Long Term Debt/Total Assets 24.83** 20.39
(12.12) (12.96) Total Assets -1.601 0.108 (2.262) (3.112) #banks -0.097 -0.106 (0.099) (0.106) Duration 0.283*** 0.275*** (0.0973) (0.104) Constant 25.27*** 26.33 10.26 (2.024) (17.58) (23.86)
Entity fixed effects included Yes Yes Yes
Year fixed effects included No No Yes
N 1061 1013 1013
Standard errors in parentheses * p<0.10, ** p<0.05, *** p<0.01
Table VII Random Effects
Independent variable: Turnover; performance measured at renegotiation year (R0)
Fixed Effects Random Effects
(1) (2)
EBITDA/Total Assets -36.05** -27.51***
(14.91) (6.546)
Interest Coverage Ratio -0.013 0.0136
(0.130) (0.011)
Long Term Debt/Total Assets 24.83** 4.296
(12.12) (3.984) Total Assets -1.601 -0.175 (2.262) (0.472) #banks -0.097 -0.039 (0.099) (0.036) Duration 0.283*** 0.309*** (0.0973) (0.046) Constant 26.33 18.66*** (17.58) (3.904) N 1013 1013
25
4.1.2 Subsample estimates
This paper also applies the regression analysis to subsamples of loans issued before or after the beginning of the financial crisis. The full sample OLS regressions are repeated including the same control variables. Table VIII, Panel A on page 27 documents the estimates for the model for loans issued between 1995 and 2007 and Panel B on page 28 documents the estimates for the model for loans issued between 2008 and 2013. Two main differences are observed in the findings; the first difference is in the magnitude of economic impact and the second difference is in the timing of performance measurement.
The economic impact of a 10% lower EBITDA/Total measured at the year of renegotiation increases turnover with 3.0 percentage points for loans issued before 2008, relative to 3.1 percentage points for the full sample. For loans issued in 2008 or later, this increases to a 4.4 percentage points higher turnover. For Interest Coverage Ratio the results indicate a different effect. Pre-crisis, a 10% lower Interest Coverage Ratio measured at the year of renegotiation increases turnover with 0.01 percentage points, in line with the full sample result, after the crisis the coefficient is no longer significant.
For the full sample, impact of the performance variables increased between origination and renegotiation. For pre-crisis estimates I found nearly the same results. Now, the coefficient for performance at the year of issuance is no longer significant and increases for the year after issuance and the year of renegotiation. For the post-crisis estimates a different pattern is observed. The effect of EBITDA/Total Assets on loan sales is the greatest when measured in the issue year. A 10% lower EBITDA/Total Assets increases turnover with 5.0 percentage points. This decreases to 4.1 percentage points a year later and to 4.4 percentage points in the year of renegotiation. Again, performance measured a year after renegotiation is not of significant influence for loan sales.
To summarize, the relationship between firm performance and loan sales described throughout this paper is of the greatest impact for loans issued after 2008. A possible explanation for this stronger impact can be lower market liquidity. After the crisis, the primary market for syndicated loans fell with 47% (Ivashina & Scharfstein, 2010) and turnover in my sample decreased from an average of 21% before 2008 to an average of 18% after 2008. In such a constrained market banks should optimize their loan sales decision and should be more selective in which loan to sell. Irani et al. (2014) also find indicators that the loan sale decision has changed, in their empirical research the relationship between loan sales
26
and reliance on wholesale funding can only be proved for the period after the crisis. The effect of performance measured in the year of issuance can be explained by the fact that banks have become more careful and want to react more promptly to negative performance of the borrower.
A graphical presentation of the changes over time can be found in Figure II and Figure III in the Appendix. Figure II shows the development of turnover over time, which is slightly decreasing with a drop in 2010. Figure III shows the development of time to renegotiation, duration, over time where a more strongly decreasing trend can be observed after the financial crisis. The number of observations in this subsample is relatively small. For a more adequate interpretation, more observations are required. Also, changing macroeconomic factors should be included in the analysis in order to draw meaningful conclusions on loans sales after the financial crisis.
27
Table VIII Subsample Analysis
Panel A Pre-crisis
Independent variable: turnover, performance variables measured at different times, (0) indicates issue year, (1) indicates year after issuance, (R0) indicates renegotiation year, (R1) indicates year after renegotiation; subsample all amended loans issued between 1995-2007
OLS
(1) (2) (3) (4) (5) (6) (7) (8)
EBITDA/Total Assets -17.93* -18.94* -15.80 -11.34 -10.74 -26.44** -30.36*** -7.302
(10.18) (10.22) (10.07) (10.37) (10.93) (7.743) (7.907) (9.691)
Interest Coverage Ratio 0.008 0.009 0.011 0.021 0.019 0.109*** 0.147*** 0.020
(0.036) (0.037) (0.036) (0.044) (0.045) (0.0372) (0.0402) (0.090)
Long Term Debt/Total Assets 1.078 0.365 -1.102 -0.432 1.683 3.920 3.037
(4.482) (4.509) (4.887) (5.066) (4.954) (4.977) (5.311) Total assets -0.834* -0.335 -0.023 0.040 -0.210 -0.192 0.083 (0.494) (0.530) (0.658) (0.693) (0.692) (0.691) (0.723) Duration 0.311*** 0.282*** 0.272*** 0.292*** 0.295*** 0.295*** (0.046) (0.048) (0.049) (0.049) (0.051) (0.053) #banks -0.044 -0.054 -0.049 -0.039 -0.043 -0.055 (0.039) (0.039) (0.040) (0.040) (0.040) (0.041) Constant 23.51*** 29.40*** 19.67*** 13.97** 12.67* 15.02** 14.61** 10.69 (1.474) (4.266) (4.515) (6.407) (7.051) (6.840) (6.816) (7.247)
Categorical variable "Rating" No No No Yes Yes Yes Yes Yes
Categorical variable "GIC" No No No No Yes Yes Yes Yes
Performance measure at T= 0 0 0 0 0 1 R0 R1
28
Panel B Post-crisis
Independent variable: turnover, performance variables measured at different times, (0) indicates issue year, (1) indicates year after issuance, (R0) indicates renegotiation year, (R1) indicates year after renegotiation; subsample all amended loans issued between 2008-2013
OLS
(1) (2) (3) (4) (5) (6) (7) (8)
EBITDA/Total Assets -32.77** -35.18** -45.48*** -44.45** -50.33** -41.30** -44.03*** -36.98
(13.92) (14.40) (15.18) (19.42) (20.50) (18.25) (16.69) (22.99)
Interest Coverage Ratio -0.017 -0.015 -0.013 0.027 0.022 0.014 -0.002 -0.036
(0.014) (0.015) (0.015) (0.151) (0.161) (0.088) (0.086) (0.171)
Long Term Debt/Total Assets -0.001 -0.045 2.738 4.988 8.539 7.423 7.804
(8.569) (8.808) (10.99) (12.03) (11.55) (11.59) (13.14) Total assets 0.987 0.467 1.166 1.769 1.849 1.871 1.704 (0.941) (0.990) (1.316) (1.452) (1.479) (1.459) (1.509) Duration 0.024 -0.114 -0.081 -0.079 -0.085 0.019 (0.126) (0.134) (0.163) (0.165) (0.176) (0.199) #banks 0.131 0.107 0.094 0.077 0.065 0.081 (0.119) (0.129) (0.137) (0.138) (0.138) (0.144) Constant 21.82** 14.08* 17.30** 8.027 4.229 0.142 0.862 1.442 (1.997) (7.702) (8.157) (12.79) (14.14) (14.33) (14.21) (14.84)
Categorical variable "Rating" No No No Yes Yes Yes Yes Yes
Categorical variable "GIC" No No No No Yes Yes Yes Yes
Performance measure at T= 0 0 0 0 0 1 R0 R1
N 209 209 208 171 171 171 169 165
Standard errors in parentheses * p<0.10, ** p<0.05, *** p<0.01
29
4.2 Robustness checks 4.2.1 Sample selection
First, the loans used in this empirical research are all renegotiated before maturity which might make the sample different from syndicated loans in general and could lead to a selection issue. Roberts and Sufi (2009) found that 96% of their sample of long term loans of US publicly listed borrowers is renegotiated before maturity. Based on this research it is fairly valid that the results of this paper apply to all syndicated loans, however a side note on the possible differences follows.
Renegotiation of the loan contract requires unanimity of all participants in the syndicate, the more participants in the syndicate, the more veto-rights to prevent renegotiation. Thus, the larger the syndicate, the harder to renegotiate the contract. From the banks’ point of view this power is a way to reduce the chance of probability of strategic default (Sufi, 2005). The number of banks in the syndicate increases in the amended sample compared to the original sample, which could be an indicator of “participant-loading” and also lead to a selection issue. This suggests that the loans in the sample on average have a perceived increased risk at renegotiation relative to origination. The sample in this paper therefore could include worse performing firms relative to syndicated loans in general. On the other hand, the fact that renegotiation on itself has taken place is a positive indicator relative to syndicated loans in general. Existing research shows that investors also value renegotiation of loan contracts positively, as concluded from the increased stock price (Gande & Saunders, 2003). Based on the available information, it is difficult to say which interpretation of the increase in the number of banks is correct. Public availability of all syndicated loan details in credit register databases would solve for this problem. In the future, loan terms such as restricting covenants should be investigated to conclude on this issue. For this paper, the evidence provided by Roberts & Sufi (2009) validates some generalization.
4.2.2 Other robustness checks
For conclusions on the consistency of estimated coefficients, the model is tested for omitted variable bias. The Ramsey RESET test is performed for the models estimated in columns (5) to (8) of the full sample analysis. The null hypothesis is that the model has no omitted variables. At a 1% significance level I failed to reject the null hypothesis for model (5), (6) and (8), so there is no evidence that additional control variables are needed. For column (7),
30
the null hypothesis should be rejected and this provides evidence for the presence of omitted variable bias in the model.
Exclusion of lender characteristics in the regression analysis is my main concern for interpretation of the results. Existing research has extensively examined the loan sale decision motivated by bank characteristics and shows evidence that those characteristics indeed explain loan sales. Irani et al. (2014) find that banks reliant on wholesale funding show higher turnover. The one available lender characteristic in my data set is an indicator variable for whether the bank is lead arranger in the syndicate or not. I have performed an additional T-test to see the difference in turnover between lead arrangers and participant lenders. The mean turnover of lead arrangers is 18.1% and the mean turnover of participant lenders is 22.4%. With a P-value of 0.001 these are significantly different. This means that a participant lender is more likely to leave the syndicate than a lead arranger. When I include the lead arranger variable in the regression it has a relationship with turnover at a 1% significance level but it does not affect the other results. In this paper, other bank characteristics are ignored but could have a relation with the turnover variable as well. In order to solve for the problem of exclusion of lender characteristics banks in the syndicates should be identified and corresponding characteristics should be included in the regression analysis.
As a robustness check for the measurement of performance variables, the models are re-estimated for the increase (decrease) in EBITDA/Total Assets and the increase (decrease) in Interest Coverage Ratio. When substituting these two variables for the increases (or decreases) in the regression analysis, the results remain the same. Therefore it is valid to assume that the results are not influenced by the way accounting performance is measured. Credit rating is a more static performance measure, which is not updated every year. In this paper the credit rating closest after issuance of the loan is used for regression analysis. In general, no major credit rating changes were observed in the sample. The ratings are performed by rating agency Standard & Poor’s, the way they evaluate the loans is the same for all firms and it is a widely accepted rating scale in the financial industry. Therefore it is valid to assume this rating is a good measure of creditworthiness.
Furthermore, this paper defines turnover as % of ownership changed between origination and renegotiation only, which might be problematic for validity of the results as well. First this might be problematic because the number of trades between the two dates is not taken into account. It could be possible that one bank has sold and bought back the loan share between
31
the two dates but this is not included in the turnover variable. Second, a percentage of number of banks that change ownership might be problematic. Different participants can hold different amounts of the loan. Now there is no distinction between a loan sale of $1 million or $100 million. The size of the loan share compared to the complete syndicated loan or the size of the bank should be taken into account. Again, availability of more detailed data on syndicated loans could solve for this problem.
4.2.3 Statistical tests
In Table IX the results of the Modified Wald Test for heteroscedasticity are presented. Heteroscedasticity is present in this sample and therefore heteroskedastic standard errors have to be obtained in regression analysis. The Woolridge Test for autocorrelation in panel data is performed and results are presented in Table IX. The result indicates that the data does not have first-order autocorrelation.
Table IX Additional tests
Modified Wald test for heteroscedasticity
Chi2= 2.60E+36
P>Chi2= 0.000
Woolridge test for autocorrelation in panel data
F= 14.91
P>F= 0.003
5. Conclusion and future research
My hypothesis based on existing research was that banks i) want to sell the loans of badly performing firms and ii) exploit a negative private information advantage in loan sales. Theory on the loan sale decision, renegotiation, bank funding, agency problems, information asymmetry and moral hazard is studied. First I find that there is a difference in number of banks in the syndicate between the time of origination and amendment. The number of banks per syndicate increases from 18 at origination to 23 at renegotiation.
32
I found some evidence for a negative relationship between firm performance and loan with OLS regressions, fixed effects regressions and random effects regressions. The results show 1) a lower EBITDA/Total Assets leading to a higher turnover within the syndicate 2) a negative relationship between Interest Coverage Ratio and turnover, and 3) that credit rating scores are of significant influence on the loan sale decision. These findings suggest that banks do want to sell the loans of badly performing companies, partially affirming the hypothesis. The economic impact of EBITDA/Total Assets on turnover is the greatest when turnover is measured in the year of renegotiation. A 10% lower EBITDA/Total Assets is associated with a 3.1 percentage points higher turnover. The findings do not indicate a relationship between performance after renegotiation and loans, which suggests that there is no evidence of banks exploiting a private negative information advantage. Therefore, the second part of the hypothesis cannot be affirmed.
Results of the pre- and post-crisis subsamples indicate that the magnitude of the EBITDA/Total Assets impact is greater for loans issued after 2008 than loans issued before 2008. For the period before 2008 performance measured in the year of renegotiation has the greatest impact, where for the period after 2008 performance measured in the issue year is of the greatest significant impact. The latter indicates a 5.0 percentage points higher turnover at a 10% lower EBITDA/Total Assets.
Future research could address the validity issues as discussed in the previous section. The possibilities are mainly dependent on the availability of data. An interesting subject expanding the results of this paper is the effect of the lender. Lender characteristics might explain more about the loan sale decision. It could also explain which banks decide to leave a syndicate, besides the explanation of which loans are sold based on borrowers’ characteristics. Another idea for future research is to investigate the differences between syndicated loan sales by banks versus other financial institutions.
